Ecosystem Services 23 (2017) 174208

Contents lists available at ScienceDirect

Ecosystem Services
journal homepage: www.elsevier.com/locate/ecoser

The eect of forest owner decision-making, climatic change and societal MARK
demands on land-use change and ecosystem service provision in Sweden

Victor Blancoa, , Sascha Holzhauera, Calum Browna, Fredrik Lagergrenb, Gregor Vulturiusa,c,
Mats Lindeskogb, Mark D.A. Rounsevella
a
Institute of Geography and the Lived Environment, School of Geosciences, University of Edinburgh, Drummond Street, Edinburgh EH8 9XP, United
Kingdom
b
Department of Physical Geography and Ecosystems Science, Geobiosphere Science Centre, Geocentrum II, Lund University, SE-223 62 Lund, Sweden
c
Stockholm Environment Institute, Linngatan 87D, SE-104 51 Stockholm, Sweden

A R T I C L E I N F O A BS T RAC T

Keywords: The uncertain eects of climatic change and changing demands for ecosystem services on the distribution of
Agent-based modelling forests and their levels of service provision require assessments of future land-use change, ecosystem service
Behaviour provision, and how ecosystem service demands may be met. We present CRAFTY-Sweden, an agent-based,
Ecosystem services supply and demand land-use model that incorporates land owner behaviour and decision-making in modelling future ecosystem
Forestry sector
service provision in the Swedish forestry sector. Future changes were simulated under scenarios of socio-
Scenario analysis
economic and climatic change between 2010 and 2100. The simulations indicate that the inuence of climatic
change (on land productivities) may be less important than that of socio-economic change or behavioural
dierences. Simulations further demonstrate that the variability in land owner and societal behaviour has a
substantial role in determining the direction and impact of land-use change. The results indicate a sizeable
increase in timber harvesting in coming decades, which together with a substantial decoupling between supply
and demand for forest ecosystem services highlights the challenge of continuously meeting demands for
ecosystem services over long periods of time. There is a clear need for model applications of this kind to better
understand the variation in ecosystem service provision in the forestry sector, and other associated land-use
changes.

1. Introduction future demands for ES supply.

Land-use and land management change have important eects on
the provision of ecosystem services (ES) (MEA, 2005). Forests provide
a particularly wide range of ES, including timber and non-timber
products, air purication, carbon sequestration, biodiversity preserva-
tion and recreation, which make fundamental contributions to human
societies and natural systems (De Groot et al., 2010; MEA, 2005).
Meanwhile, pressures on the world's forests are increasing. Clearance
for agriculture and timber harvesting, preservation and planting for
climate mitigation, and climate-driven changes in growing conditions
are all likely to interact to transform future ES provision (Buonocore
et al., 2012; Zanchi et al., 2012; Alexandratos and Bruinsma, 2012;
Schroter et al., 2005; Soja et al., 2007; Tilman et al., 2001). Hence,
forest management strategies are being revised (e.g. Jonsson et al.,
2015; Kjaer et al., 2014) and future land-use change assessed (e.g.
Thompson et al., 2011) through the use of computer models in an
attempt to support adaptation to changing conditions and to meet
One of the diculties in designing management strategies for
future conditions is the need to anticipate demands for ES. Such
demands are dicult to estimate (Hayha et al., 2015), and forest
modelling generally focuses only on timber yields as a result (Brown
et al. in press). Where a wider range of ES are considered, assessment
is often based on maps of suitability (e.g. Hayha et al., 2015; Sohel
et al., 2015) or vulnerability (e.g. Metzger et al., 2008; Tzilivakis et al.,
2015), which need not consider ES demand. Mapping of ES supply is
also performed through ES valuation (e.g. Costanza et al., 1997), which
assumes demands non-explicitly. Even where ES demands are ac-
knowledged, only services with a market value are included (e.g.
Verkerk et al., 2014). As a result, no study has investigated the
provision of non-marketable ES in relation to demand levels; a
necessary step to identify likely and desirable changes that enable
forests to satisfy societal needs for ES.
Another challenge faced by models of future ES provision is that
land use and management change, which determine ES provision,

Corresponding author.
E-mail address: v.blanco@ed.ac.uk (V. Blanco).

http://dx.doi.org/10.1016/j.ecoser.2016.12.003
Received 17 May 2016; Received in revised form 22 November 2016; Accepted 7 December 2016
Available online 03 January 2017
2212-0416/ 2016 Elsevier B.V. All rights reserved.
V. Blanco et al.
ultimately depend on the decisions of land owners. In forestry,
behavioural and cognitive factors such as owner objectives and
attitudes are known to have strong inuences on management choices
(Andersson and Gong, 2010; Ingemarson et al., 2006; Vulturius et al.
in review). However, due to lack of data and the uncertainty associated
with long time horizons in forest management they are seldom
incorporated into models, and never at scales larger than individual
landscapes (Blennow et al., 2014; Rammer and Seidl, 2015; Brown at
al. 2016). Nevertheless, in terms of potential responses to future
change owners can eectively be distinguished by the denition of
categories (Blanco et al., 2015; Karali et al., 2013). Such categorisa-
tions are particularly useful in agent-based modelling (ABM) of land-
use change, allowing the decision-making of individual land managers
to be simulated heterogeneously and across large geographical extents
(Matthews et al., 2007; Valbuena et al., 2010). The adoption of such
models to map the eects of human behaviour on ES is recent (Boone
et al., 2011; Brown et al., 2014; Murray-Rust et al., 2011), but holds
great promise to reect the drivers and consequences of change in the
forestry sector more accurately.
The need for improved modelling of the forest sector is clearly
illustrated by countries such as Sweden, which have large forest areas
that are economically and culturally important, and which are likely to
be signicantly aected by climatic change. Sweden has 69% forest
cover (SLU, 2015), of which approximately 50% is owned by individual
owners (Swedish Forest Agency, 2015) with diverse objectives. In
2011, forestry accounted for 2.2% of Swedish GDP. We therefore adopt
Sweden as a case study for the development of a forest management
ABM that accounts for land owner decision-making and is capable of
appraising provision of a wide range of ES under projected future levels
of demand. We apply this model at national scale under combined
socio-economic and climatic scenarios (Shared Socio-economic
Pathways and Representative Concentration Pathways SSPs-RCPs;
O'Neill et al., 2014, van Vuuren et al., 2011 2014) from 2010 to 2100.
The purpose of this exercise was to explore: a) future ES provision and
how ES demands may be met, b) land-use change, and c) changes in
land owner objectives, in the Swedish forestry sector.
2. Methodology
To explore future ES provision and land-use change in Sweden,
Fig. 1. Schematic representation of the structure of the CRAFTY-Sweden model showing ows (solid lines) and associations (dashed lines) between components.
175
Ecosystem Services 23 (2017) 174208
focusing on forestry, we developed the CRAFTY-Sweden model, based
on the CRAFTY agent-based modelling framework (Murray-Rust et al.,
2014) (see Appendix A for the model ODD protocol). CRAFTY allows
the representation of large-scale land-use dynamics, based on demand
and supply of ES (e.g. timber, food). Demand is given exogenously
while supply depends on the productivities and behaviours of modelled
agents, and the productivities of agents locations (described by capitals
representing the availability of resources such as infrastructure, human
capital and crop suitability). Geographical space is represented as a
grid of cells, each of which has dened levels of a range of capitals.
Each cell may be managed by a single land-use agent, which uses the
capital stock available within the cell to provide services according to
its own production function. The competitiveness of a given level of
service provision can be calculated on the basis of societal demands,
overall supply levels and benet functions, which describe the
monetary and non-monetary value to society of service production.
Agents make decisions based on their current competitiveness and
participate in an allocation procedure with potential new agents that
may result in land-use change. We use agent functional types (Arneth
et al., 2014; Rounsevell et al., 2012) (hereafter agent types) for the
denition of agent production and behaviour. This approach helps to
characterise agent typologies that dene general characteristics of
agents, from which individual agents can subsequently be drawn.
2.1. Model description
In CRAFTY-Sweden, agents include dierent types of forest owners
and farmers. Farmers were dened to simulate the competition for land
between forestry and agriculture. Forest owner decision-making in-
volves four key components: 1) owner objectives and associated
management practices, 2) the time of felling, 3) an estimation of the
future benets agents expect to obtain from their land-use, 4) and their
willingness to abandon, change management or hand over land to a
dierent owner considering their competitiveness. Farmers consider all
but the second component. Using land productivities and infrastruc-
ture, modelled forest owners are able to produce timber from dierent
tree species, carbon sequestration, biodiversity, and recreation, while
modelled farmers choose to produce one or more services among
cereal, meat and recreation (Fig. 1).
In the following, we describe the development of the land owner
V. Blanco et al. Ecosystem Services 23 (2017) 174208

typology, and the owner production and decision-making mechanisms.
We also explain how baseline capitals, land-use and land owner types
were mapped throughout Sweden. Finally, we describe the approach to
scenarios and the analysis of simulation results.
2.1.1. Land owner typology: design and validation
We developed a typology of forest and agricultural agents, focusing
especially on the former. To dene forest owner types we used as a
basis the forest owner typology described by Blanco et al. (2015).
Because this typology was based on studies performed at dierent
scales and contexts to those found in Sweden, we performed a
validation exercise of the typology using empirical information from
872 Swedish forest owners (Vulturius et al. in review). A cluster
analysis showed that the ve overarching management roles identied
by the theoretical typology (productionist, multi-objective, recreation-
alist, conservationist and passive) were also clearly discernible in the
empirical data. Supplementary materials on this validation can be
found in Appendix B.1.
Within each overarching management role, dierent options for
forest management are possible, including the use of dierent types of
forest (dened by species composition). Forest types were assigned to
each management role on the basis of existing forest stand composi-
tions (Swedish Forest Agency, 2015) and potential adaptation mea-
sures to climate change that consider species composition, number of
thinnings and rotation lengths (Felton et al., 2016; Jonsson et al.,
2015) on the basis of owner objectives (Blanco et al., 2015; Duncker
et al., 2012a). Forest types assigned were pine (Pinus sylvestris),
spruce (Picea abies), boreal broadleaf (Betula pendula, B. pubescens,
Alnus incana, A. glutinosa, Populus tremula), nemoral broadleaf
(Fagus sylvatica, Quercus robur, Fraxinus excelsior, Ulmus glabra,
Tilia cordata, Carpinus betula), and combinations of these, resulting
in 17 forest owner types. The management and decision-making
strategies of each owner type are described in Sections 2.1.32.1.5.
Given the current levels of agricultural production (Swedish Board
of Agriculture, 2009) and management intensities prevailing in Sweden
(Institute of Environmental Studies, 2015), farmers were separated by
the main services provided (i.e. cereal or meat) in combination with
their main objectives (i.e. commercial or non-commercial).
2.1.2. Capitals
The capitals that agents can use in service production are produc-
tivities for pine, spruce, boreal broadleaf, and nemoral broadleaf
forests, grassland productivity (natural capital), and transportation
infrastructure (infrastructure capital). Table 1 shows capital descrip-
tions, their data sources, and the ES they contribute to producing (see
Appendix B.2 for further detail on the calculations that led to nal
capitals).
2.1.3. Baseline land-use and land owner distribution
2.1.3.1. Land-use map. To create a baseline land ownership map for
2010 we rst devised a land-use map at 1 km2 resolution that included
pine, spruce, pine-spruce, pine-boreal broadleaf, spruce-boreal
broadleaf, boreal broadleaf, and nemoral broadleaf productive
forests, agriculture, protected areas, non-productive forests, semi-
natural vegetation, wetlands, open spaces, other unmanaged land,
articial, and water bodies. SLU Forest Map data (SLU, 2010) on the
proportion of dierent tree species per cell were used to identify forest
cover and classify it according to the forest types assigned above,
according to the proportion of forest within the cell, and the
proportions of dierent species within that forest. CORINE land
cover (EEA, 2014) was used to identify all other land use/land cover
classes. Nationally Designated Areas (EEA, 2015) were then
superimposed to dene protected areas. Non-productive forests are
also protected and unavailable for production (Swedish Forest Agency,
2014). Thus, we identied them by:
1. Assigning to forested cells the value of the highest productivity found
among all forest types within that cell; and
2. Given the proportion of non-productive forest per county (Swedish
Forest Agency, 2015), selecting for each county the equivalent
number of cells with the lowest productivity values.
Mean forest age values from the SLU Forest Map were used to
assign forest ages. While this dataset allows mapping of forest age
throughout the country, it was found to overestimate the area covered
by middle aged forest, planted during 19301970, and underestimate
that of old forest and very young forest (see Fig. B.2), when compared
to age frequency distribution in the National Forest Inventory (Swedish
Forest Agency, 2015).
2.1.3.2. Agent locations. Forest owner types were allocated to
productive forest types using data about: a) the area of productive
forest land by county and ownership class for 2010 (Swedish Forest
Agency, 2015); and b) the proportion of owners in each county
belonging to each group from the cluster analysis. Agricultural land
and (some) semi-natural vegetation were assigned to commercial

Table 1
Identities and data sources for modelled capitals, and the ecosystem services they contribute to producing.

Capital Denition Input data Ecosystem services Data source

(units; resolution)

Pine, Spruce, Boreal Broadleaf, and
Nemoral Broadleaf Forest
Productivities
Grassland Productivity
Transportation Infrastructure
Baseline productive potential Forest production potential per forest type 1. Pine, spruce, boreal br. and (Hgglund and
for each forest type nemoral br. timber Lundmark, 1987)
3 -1 -1 2
(m sk ha yr ; 1 km ) 1. Carbon (Johansson et al. 2013)
2. Biodiversity SLU
Baseline productive potential LPJ-GUESS simulated C3-grass NPP 1. Meat Simulations done for this
for grassland and cropland projection for 2010 driven by climate 2. Cereal article (seeSection 2.2)
(radiation, temperature, precipitation)
(kg C m2 yr1; 5050 km)
Proximity to transportation 1. Road and rail networks - Pine, spruce, boreal br. and 1. UNECE
networks and central markets 2. Waterway networks nemoral br. timber 2. EEA
3. Travel time to nearest town with over 1. Meat 3. GEMU, JRC
50,000 inhabitants (1 km2) 2. Cereal
3. Recreation

SLU: SLU Forest Map, produced by Swedish University of Agricultural Sciences, accessed via ftp://salix.slu.se/download/skogskarta
UNECE: United Nations Economic Commission for Europe, accessed via http://www.unece.org/trans/areas-of-work/transport-statistics/statistics-and-data-online.html
EEA: European Environment Agency, accessed via http://www.eea.europa.eu/data-and-maps
GEMU, JRC: Global Environment Monitoring Unit, managed by the Joint Research Centre, accessed via http://www.edenextdata.com/?q=content/jrc-accessibility-map-estimated-
travel-time-nearest-city-population-50000

176
V. Blanco et al. Ecosystem Services 23 (2017) 174208

Fig. 2. Map of land owner type distribution throughout Sweden in 2010. Available and unavailable land refer to land that can or cannot be managed by agents, respectively.

Table 2
Percentage of land area occupied by each owner type, per forest/ farmland type and management role, in Sweden in 2010. Percentages for forest owner types are given out of the total
productive forest land, and out of total farmland for farmers.

Forest Owner Type Productionist Multi-objective Recreationalist Conservationist Passive Total

Pine 8.34 8.34

Spruce 0.60 0.60
Pine-Spruce 30.00 22.50 13.78 66.28
Pine-Boreal Broadleaf 10.37 0.86 11.23
Spruce-Boreal Broadleaf 10.74 2.50 13.24
Boreal Broadleaf < 0.01 0.01 < 0.01 < 0.01 < 0.01 0.02
Nemoral Broadleaf 0.17 0.08 0.03 0.28
Total 38.94 43.62 13.95 0.08 3.39 100
Farmland Owner Type Commercial Non-commercial Total
Cereal 86.86 5.02 91.88
Livestock 3.94 4.17 8.11
Total 90.80 9.19 100

cereal, non-commercial cereal, commercial livestock, and non-
commercial livestock farmer agents according to the land-use
intensity in 2010 (Institute of Environmental Studies, 2015).
Remaining semi-natural vegetation, wetlands, and other unmanaged
land were left unallocated. Protected areas, non-productive forests,
open spaces with little or no vegetation, articial surfaces, and water
bodies were not available for allocation during simulations. Fig. 2
shows the resulting map of Sweden, and Table 2 presents the share of
land occupied by each owner type. Further detail on the creation of
land-use and land owner type maps can be found in Appendix B.3.
2.1.4. Land owner service production
The production of agricultural services was modelled on an annual
basis. Forestry services are however dependent on forest age.
Additionally, climatic change can aect service production by acting
on productivities. We therefore developed ways of modelling the time-
dependent component of the dierent services in CRAFTY. The
production of a service by an agent in a given year was based on the
Cobb Douglas function, adapted to incorporate a time component (Eq.
(1)).
ps = os, a, t c (ci +ci,t )c,a (1)
Production ( ps ) of a service s by an agent a within a cell at time t
depends upon the initial (unit-less, i.e., [0-1]) levels of capitals (c )
within that cell plus annual climate-induced change in cell capitals
(ci +ci, t ), weighted by the capital sensitivities of the agent type (c, a )
and multiplied by the optimal production potential of that agent type
(os, a, t ) (Table 3). The optimal service production os, a, t is the maximum
production achievable in Sweden under ideal environmental condi-

177
 Table 3
Land owner type production, felling age and competitiveness (scenario-independent) parameters. Number of stems planted per ha for each forest type, site index, number of thinnings implemented, age at each thinning per forest type, and
percentage removed per thinning are parameters given to ProdMod to calculate the (age-dependent) optimal timber production and (above ground) carbon sequestration functions. Additional information on ProdMod simulations and parameters
can be found in Appendix B.4. These functions and remaining parameters in this table are CRAFTY-Sweden inputs. Yearly os illustrates yearly optimal farmer production. Productivity and infrastructure sensitivities ( ) are given per service. Felling
V. Blanco et al.

age means ( ) and standard deviations ( ) represent number of years past minimum felling age (m.f.a.) of a forest, given as the m.f.a. range dependent on site quality.

Service Production Felling Age Competitiveness

Land Owner Type No. Stems/ha Site Index dm No. Thinnings Age at each Thinning % Removed per Thinning Yearlyos m.f.a. , Years Probability

(tonne) Productiv. Infrastr. Range past m.f.a. Giving-up

a e f a g
Productionist 2500 280 3 24, 39, 54 25, 20, 20 - 0.8 , 1 , 0.06 0.2 , 1 65100 12, 10 0.05
Pine
Productionist 2600 360 3 23, 38, 53 25, 20, 20 0.8b, 1e, 0.06f 0.2b, 1g 4595 10, 8 0.05
Spruce
Productionist 1250, 1300 280, 360 3 24/23, 39/38, 54/53 25, 20, 20 0.8a,b, 1e, 0.06f 0.2a,b, 1g 4595 10, 8 0.05
Pine-Spruce
c e f c g
Productionist 2200 320 3 15, 30, 45 25, 20, 20 0.8 , 1 , 0.06 0.2 , 1 4060 9, 7 0.05
Boreal Br.
a,b e f a,b g
Multi-objective 1150, 1250 280, 360 2 24/23, 39/38 30, 25 0.85 , 1 , 0.06 0.1 , 0.8 4595 15, 12 0.05
Pine-Spruce
Multi-objective 1840, 420 280, 320 2 24/20, 39/35 25/50, 20/25 0.85a,c, 1e, 0.06f 0.1a,c, 0.8g 65100 10, 8 0.05
Pine-Boreal Br.
Multi-objective 2000, 420 360, 320 2 23/20, 38/35 25/45, 20/25 0.85b,c, 1e, 0.06f 0.1b,c, 0.8g 4595 10, 8 0.05
Spruce-Boreal Br.
c e f c g
Multi-objective 2100 320 2 15, 30 30, 25 0.85 , 1 , 0.06 0.1 , 0.8 4060 15, 12 0.05
Boreal Br.
Recreationalist 1100, 1100 280, 360 3 24/23, 39/38, 54/53 25, 20, 20 0.9a,b, 1e, 0.06f 0.3a,b, 0.6g 4595 80, 14 0.05
Pine-Spruce

178
Recreationalist 2000 320 3 15, 30, 45 25, 20, 20 0.9c, 1e, 0.06f 0.3c, 0.6g 4060 100, 14 0.05
Boreal Br.
d e f d g
Recreationalist 1250, 1250 350, 300 3 25/22, 40/37, 55/52 25, 20, 20 0.9 , 1 , 0.06 0.3 , 0.6 110150 60, 14 0.05
Nemoral Br.
c e f c g
Conservationist 2100 320 1 15 35 0.9 , 1 , 0.06 0.3 , 0.8 4060 100, 14 0.05
Boreal Br.
Conservationist 1250, 1250 350, 300 1 25/22 35 0.9d, 1e, 0.06f 0.3d, 0.8g 110150 60, 14 0.05
Nemoral Br.
Passive 0.9a,c, 1e, 0.06f 0.1a,c, 1g 65100 25, 17 0.05
Pine-Boreal Br.
b,c e f b,c g
Passive 0.9 , 1 , 0.06 0.1 , 1 4595 25, 17 0.05
Spruce-Boreal Br.
Passive 0.9c, 1e, 0.06f 0.1c, 1g 4060 15, 10 0.05
Boreal Br.
Passive 0.9d, 1e, 0.06f 0.1d, 1g 110150 10, 10 0.05
Nemoral Br.
h h
Commercial 201 0.8 0.5 0.2
Cereal
h g,h
Non-commercial 121 0.5 0.3 0.2
Cereal
Commercial 324 0.6i 0.5i 0.2
Livestock
(continued on next page)
Ecosystem Services 23 (2017) 174208
V. Blanco et al. Ecosystem Services 23 (2017) 174208

Competitiveness tions, and its levels are taken from an agent type-specic function (i.e.
optimal production function) that describes the change in service
Probability
production levels with forest age. Capital (i.e. land productivity and
Giving-up infrastructure) levels are used to adjust optimal production to produc-
tion levels achievable at a particular site. Capital sensitivities adjust
0.2 these site-specic production levels to what the agent type would be
able to attain given its particular dependencies on capitals. To reect
past m.f.a.
, Years

individual variability, each agent's optimal production value and capital

sensitivity levels were randomly drawn from uniform distributions
around the agent type's mean values ([0.95os, t ,1.05os, t ],

[c 0.1,c +0.1]). Timber supply (i.e. harvest) is recorded upon felling,

Felling Age

while other forest services are supplied yearly. Production calculations

Range

for each service are described below.

m.f.a.

2.1.4.1. Timber. For timber production, ost is given by a forest owner

Infrastr.

type-specic function that determines timber growth given the forest's

age. The ProdMod model (Eko, 1985) was used to generate timber
0.2 g,i

growth curves for each owner type given their management

preferences. Given passive owners generalised lack of primary

objectives for forestry, we assumed them to inherit forest, and
Productiv.

therefore only enabled them to take over the forest and associated
optimal production function of other owner types managing forests
0.3i

with the same tree species as them. Hence, optimal production

functions were not calculated for passive owners. Table 3 shows

Yearlyos

(tonne)

parameter values used in ProdMod that diered for each owner type.
See Appendix B.4.1 for further detail on optimal timber production
193

function calculation.
% Removed per Thinning

2.1.4.2. Carbon sequestration. Due to the diculty of calculating soil

carbon levels in interaction with forest productivities, only above-
ground sequestered carbon (excluding the stump) was calculated.
Optimal production functions of above ground carbon were also
calculated using ProdMod outputs (Appendix B.4.2).

Age at each Thinning

2.1.4.3. Biodiversity. The calculation of optimal forest biodiversity

production considered forest age (Duncker et al., 2012b; Koskela et al.,
2007; Marchetti, 2004), using the generation of coarse woody debris
with age as a proxy (e.g. (Berg et al., 1994; Jonsell et al., 1998;
Siitonen, 2001), tree diversity (Gamfeldt et al., 2013; Marchetti, 2004)
and management practices undertaken by each owner type (e.g. woody

debris removal), which have an inuence on biodiversity (Blanco et al.,

No. Thinnings

2015; Duncker et al., 2012a, 2012b). We chose these forest attributes

as indicators of biodiversity because of the availability of baseline data
for them and the possibility of updating the data during model
simulations. Finally, we considered the eect of forest productivity
Service Production

on biodiversity, specically on coarse woody debris (Sturtevant et al.,

1997), by assigning sensitivities to timber productivities. For further

Site Index dm

details of the calculation of optimal biodiversity production functions

see Appendix B.4.3.

2.1.4.4. Recreation. Recreational value in Scandinavia is largely

No. Stems/ha

determined by the age of a forest, but also by forest management

practices, accessibility and, to a lesser extent, by the types of tree
species present (i.e. conifer vs broadleaf, and monoculture vs mixed)
nemoral broadleaf timber.

(Edwards et al., 2012). See Appendix B.4.4 for further detail on optimal

boreal broadleaf timber.

recreation function calculation.

carbon sequestration.
Land Owner Type
Table 3 (continued)

spruce timber.
Non-commercial

Pine timber.

biodiversity.
recreation.

2.1.4.5. Cereal and meat. Given baseline maps with available capitals
Livestock

and commercial cereal, non-commercial cereal, commercial livestock

cereal.
meat.

and non-commercial livestock agent locations (see Section 2.1.6), their

os and c were adjusted until total cereal and meat production equalled
h
d
b

g
a

e
c

I
f

179
V. Blanco et al.
the total production in Sweden reported by the FAO (2015) for 2010.
The production of non-commercial agents was set at 0.6 times that of
the commercial agents to reect approximate dierences in production
potentials across equivalent classes Van Asselen and Verburg (2013a,
2013b).
2.1.5. Forest felling
The forest in a cell is clear-felled when it reaches an age that
depends on site quality (i.e. productivity) (Lagergren et al., 2012) and
owner objectives. In Sweden, the stand age at felling is regulated by law
for pine and spruce to guarantee that the production potential is
utilised (Kunskap Direkt, 2015), and for beech, birch and oak
recommended rotation periods exist (Lf et al., 2009; Rytter et al.,
2008). Hence, lowest minimum felling age was assigned to the highest
productivity values, while highest minimum felling age corresponded
to the lowest productivity values (Table 3; Appendix B.5). Also, each
owner type was assigned a Gaussian distribution of the planned felling
age (above minimum felling age) (Table 3). This distribution was
dened as being within the recommended rotation periods for all
owner types except for recreationalists, conservationists and passive
owners managing broadleaf forests. As these latter groups are not
primarily interested in timber production (Blanco et al., 2015), they
were assigned felling age distributions beyond the recommended
rotation period. Felling age is determined at the time that an agent is
allocated to a cell by randomly drawing a number (i.e. age) from within
the agent type's distribution. Upon felling, timber is harvested and
carbon sequestered in the standing timber is removed from the
national pool.
2.1.6. Competition for land
Farmers can be taken over by other agents each year because they
are assumed to manage on annual timescales. For forest owners
however, we assume that they will not abandon or change the
management approach on their land until the forest has reached
maturity, in order to recover the initial investment. Hence, competition
for forested land starts only once the minimum age of felling has been
reached. At that point a potential agent with a higher competitiveness
score than the incumbent agent can take over its land, resulting in one
of two outcomes:
1. If the potential agent is a forest owner type willing to plant the same
forest type as that already standing in the cell, it will inherit the
production functions of the former owner, as the eect of changing
management of a forest once maturity is reached is negligible. Age of
felling is however adjusted to meet the objectives of the new agent.
As mentioned in Section 2.1.3, passive owners follow this system
exclusively and do not compete for unmanaged land.
2. If the potential agent is a farmer or a forester not meeting the above
criteria, the standing forest is clear-felled and land is either
converted to farmland or to newly-planted forest.
Forest owners plan what they will plant according to (non-climate
sensitive) charts that show potential tree growth according to site
conditions. Even though some owners may also consider climate
change and risk spreading, this is currently not a generalizable trait
of Swedish forest owner decision-making (Blennow et al., 2012). They
generally use experience-based, practical knowledge, rather than
abstract theoretical knowledge about possible future developments
(Lidskog and Sjdin 2014). Hence, while farmers service production is
evaluated for the coming year, forest owners evaluate it for the (future)
year of felling, based on present (rather than future) environmental
and socio-economic conditions. To evaluate agent competitiveness for a
given bundle of services we use the benet () function (Eq. (2)):
180
Ecosystem Services 23 (2017) 174208
ps / Ai us
= bs
s
ds ds (2)
where the production level ps is time-discounted by forest age at felling
( Ai ) to reect desire for shorter-term returns where possible. Time-
discounted production is normalised by the current per-cell demand
(ds ) (i.e. demand divided by the total number of cells) to achieve levels
that are comparable across agent types supplying services that are
measured in dierent units. The per-cell unmet demand (us ), is also
normalised by the demand to give a proportional unmet demand.
Finally, bs is a weighting factor representing the importance of the
unmet demand of each service. Ai and bs are parameterised to reect
observed time discounting and the assumed importance to society of
meeting service demand levels, respectively.
If an existing agent's competitiveness is lower than its giving-up
threshold, it will abandon the cell. If a potential agent's competitive-
ness within a cell is greater than the existing agent's by a value larger
than the existing agent's giving-in threshold, then the potential agent
takes over the cell. Giving-up and giving-in thresholds reect minimum
acceptable benet and tolerance to competition respectively, and are
drawn from agent type-specic Gaussian probability distributions to
simulate individual dierences (Murray-Rust et al., 2014). Also,
because not all farmers and foresters are aected by market conditions
to the same degree or at the same time, we implement giving-up
probabilities for each agent type that apply to agents whose competi-
tiveness falls below their giving-up threshold. Giving-up and giving-in
mean values are scenario-dependent and are given in Section 2.2, while
standard deviations were set at 0.1 for all agents. Giving-up probabil-
ities are given in Table 3.
2.2. Scenario analysis
Five future scenarios were dened by combining RCPs and SSPs.
RCP4.5 (lower emission scenario) was combined with SSPs 1
(Sustainability), 3 (Regional rivalry) and 4 (Inequality), and RCP
8.5 (higher emission scenario) with SSPs 3 and 5 (Fossil-fuelled
development), so as to explore coherent combinations of emission
and socio-economic futures (Carter et al., 2015). Each RCP was also
simulated with three climate models. Each climate model-RCP combi-
nation consisted of a dierent set of climate-induced annual produc-
tivity changes. Table 4 presents scenario-specic parameters, and
scenario narratives for the dierent SSPs used here can be found in
Table B.3.
The ecosystem model LPJ-GUESS (Smith et al., 2001) was used to
simulate forest dynamics during 20102100 using climate projections
of the Global Circulation Model-Regional Circulation Model ensembles
(hereupon climate models) EC-Earth-RCA4, IPSL-RCA4 and
NorESM-RCA4 for RCPs 4.5 and 8.5 from the EURO-CORDEX project
(Jacob et al., 2014; Jones et al., 2011). Annual climate-induced change
was calculated for all productivities using LPJ-GUESS spatial projec-
tions of yearly timber volume growth for pine, spruce, boreal broadleaf
and nemoral broadleaf forests, and yearly net primary productivity
(NPP) change for grass until 2100 at 5050 km resolution. Upon
checking for non-linearities in volume growth and NPP change
projections, linear models were considered to be adequate. Therefore,
a regression coecient was calculated for every cell by performing
linear regression on projected growth values. These values were then
downscaled to 1 km2. See Appendix B.6 for more details on calculations
of climate impact on productivities.
Following land-use and European SSP storylines from Engstrom
et al. (2016) and Kok et al. (2015) respectively, SSPs diered in: a)
future demands for ES, b) probability distributions for owner type
giving-in and giving-up thresholds, c) the importance to society of
meeting demands for each service, and d) the possibility of farmland
displacing forest land. Baseline demands for timber, cereal and meat
were assumed to equal the observed production in 2010 (FAO, 2015;
V. Blanco et al. Ecosystem Services 23 (2017) 174208

Table 4
Scenario matrix. Service demands are only shown for the years 2010, 2050 and 2100. Demands shown for 2010 under the Reference scenario also apply to all other scenarios.

Reference SSP 1 - RCP 4.5 SSP 3 - RCP 4.5 SSP 3 - RCP 8.5 SSP 4 - RCP 4.5 SSP 5 - RCP 8.5

Service Demands 2010 2050 2100 2050 2100 2050 2100 2050 2100 2050 2100 2050 2100

Pine Timber 35.29 35.29 35.29 37.98 40.22 32.85 31.14 33.00 31.77 36.71 36.24 34.97 35.30
(mill. m3over bark)
Spruce Timber 44.45 44.45 44.45 47.84 50.66 41.37 49.22 41.56 40.01 46.23 45.65 44.04 44.46
(mill. m3over bark)
Boreal Br. Timber 7.64 7.64 7.64 8.22 8.71 7.11 6.74 7.14 6.88 7.95 7.85 7.57 7.64
(mill. m3over bark)
Nemoral Br. Timber 1.51 1.51 1.51 1.63 1.72 1.41 1.33 1.41 1.36 1.57 1.55 1.50 1.51
(mill. m3over bark)
Carbon Sequestration 592 592 592 637 675 551 522 554 533 616 608 587 592
(mill. tonne C)
Biodiversity 234 234 234 234 257 210 183 206 175 222 211 199 189
(1000, unitless)
Recreation 345 345 345 345 380 311 270 304 258 328 312 293 279
(1000, unitless)
Cereal 4.32 4.32 4.32 6.28 6.27 6.60 8.46 6.55 8.38 5.82 5.88 7.12 7.30
(mill. tonne)
Meat 537 537 537 765 718 808 966 815 998 761 781 1226 1178
(1000 t)
Importance of meeting bs bs bs bs bs bs
service demands
Pine Timber 1.1 1.1 1.1 1.1 1.09 1.1
Spruce Timber 1.1 1.1 1.1 1.1 1.09 1.1
Boreal Br. Timber 0.045 0.045 0.045 0.045 0.040 0.045
Nemoral Br. Timber 0.010 0.010 0.010 0.010 0.007 0.01
Carbon Sequestration 0.2 0.35 0.1 0.1 0.1 0.1
Biodiversity 1.1 2.50 1.0 1.0 1.0 1.0
Recreation 1.1 2.35 1.0 1.0 1.0 1.0
Cereal 5.0 4.5 5.5 5.5 5.5 5.5
Meat 4.5 4.0 4.8 4.8 4.5 5.0
SSP Scenario Matrix (Continue)
Reference SSP 1 - RCP 4.5 SSP 3 - RCP 4.5 SSP 3 - RCP 8.5 SSP 4 - RCP 4.5 SSP 5 - RCP 8.5
Agent Type Thresholds Give Up Give In Give Up Give In Give Up Give In Give Up Give In Give Up Give In Give Up Give In
Productionist Pine 0.30 0.10 0.30 0.10 0.28 0.12 0.28 0.12 0.28 0.12 0.30 0.10
Productionist Spr 0.30 0.10 0.30 0.10 0.28 0.12 0.28 0.12 0.28 0.12 0.30 0.10
Productionist Pin-Spr 0.30 0.10 0.30 0.10 0.28 0.12 0.28 0.12 0.28 0.12 0.30 0.10
Productionist Bor. Br. 0.26 0.10 0.26 0.10 0.24 0.12 0.24 0.12 0.24 0.12 0.26 0.10
Multi-objective Pin-Spr 0.29 0.10 0.29 0.10 0.27 0.12 0.27 0.12 0.27 0.12 0.29 0.10
Multi-objective Pin-Bor. Br 0.25 0.10 0.25 0.10 0.23 0.12 0.23 0.12 0.23 0.12 0.25 0.10
Multi-objective Spr-Bor. Br 0.25 0.10 0.25 0.10 0.23 0.12 0.23 0.12 0.23 0.12 0.25 0.10
Multi-objective Bor. Br. 0.25 0.10 0.25 0.10 0.23 0.12 0.23 0.12 0.23 0.12 0.25 0.10
Recreationalist Pin-Spr 0.24 0.30 0.24 0.30 0.22 0.32 0.22 0.32 0.22 0.32 0.24 0.30
Recreationalist Bor. Br. 0.20 0.30 0.20 0.30 0.18 0.32 0.18 0.32 0.18 0.32 0.20 0.30
Recreationalist Nem. Br. 0.20 0.30 0.20 0.30 0.18 0.32 0.18 0.32 0.18 0.32 0.20 0.30
Conservationist Bor. Br. 0.23 0.30 0.23 0.30 0.21 0.32 0.21 0.32 0.21 0.32 0.23 0.30
Conservationist Nem. Br. 0.23 0.30 0.23 0.30 0.21 0.32 0.21 0.32 0.21 0.32 0.23 0.30
Passive Pin-Bor. Br. 0.10 0.30 0.10 0.30 0.08 0.32 0.08 0.32 0.08 0.32 0.10 0.30
Passive Spr-Bor. Br. 0.10 0.30 0.10 0.30 0.08 0.32 0.08 0.32 0.08 0.32 0.10 0.30
Passive Bor. Br. 0.10 0.30 0.10 0.30 0.08 0.32 0.08 0.32 0.08 0.32 0.10 0.30
Passive Nem. Br. 0.10 0.30 0.10 0.30 0.08 0.32 0.08 0.32 0.08 0.32 0.10 0.30
Commercial Cereal 0.24 0.70 0.24 0.70 0.22 0.72 0.22 0.72 0.22 0.72 0.24 0.70
Non-commercial Cereal 0.00 0.90 0.00 0.90 0.00 0.92 0.00 0.92 0.00 0.92 0.00 0.90
Commercial Livestock 0.24 0.70 0.24 0.70 0.22 0.72 0.22 0.72 0.22 0.72 0.24 0.70
Non-Commercial Livestock 0.00 0.90 0.00 0.90 0.00 0.92 0.00 0.92 0.00 0.92 0.00 0.90
Farm can take over forest NO NO YES YES NO YES

Swedish Forest Agency, 2015), while those for carbon sequestration,
biodiversity and recreation were assumed to equal the simulated
baseline supply. Future projections were calculated using the IIASA
SSP data (IIASA, 2015) on decadal rates of change of global forest land
cover (for timber and carbon sequestration), and crop and livestock
demands. Demands for biodiversity were projected following the SSP
storylines with guidance from modelled global future changes in
species abundance from UNEP (2007). Rates of change in recreational
demands were assumed to be the same as those for biodiversity.
Giving-in thresholds were set higher and giving-up thresholds lower for
SSPs with greater barriers to adaptation (i.e. SSPs 3 and 4). See
Appendix B.7 for further details about the creation of service demand
projections.
CRAFTY-Sweden simulations were run for Sweden for the 2010
2100 period at a 1 km2 resolution. The model was calibrated to produce
minimal short-term (decadal) changes in land management under
constant levels of demand and productivities, so that the eects of long-
term forest management and scenario conditions could be isolated. The
model was then run under these static conditions (i.e. no change in
climate or demands through time) for the period 20102100 to
produce a reference scenario. To understand model behaviour, sensi-
tivity analysis was performed by altering values of behavioural, benet
function components, demands and productivities individually (results
are not reported here, but their implications are discussed in Section
4). To measure the eect of random model components 32 simulations
were run under dierent random seeds, but otherwise identical
parameterisations. Consequently, each climate model-RCP-SSP com-
bination was run once (under one random seed).

181
V. Blanco et al.
Fig. 3. Means (continuous lines) and ranges (semi-transparent areas) (dened by the three climate models) of ecosystem service provision, and service demands (dashed lines) through
the study period, under the Reference and SSP-RCP.
2.3. CRAFTY-Sweden outputs
Modelling outputs presented here are for land-use change and ES
provision. Agent types were mapped for every year during 20102100,
and are presented for each scenario grouped into land-use and
management role categories through maps depicting hotspots of
change, and gures showing nationally and regionally aggregated
change. Hotspots were dened as 5050 km (smaller next to country
borders and on islands, down to 39 km2) units of analysis where the
category with the highest proportional increase (calculated as the mean
increase from the three climate model runs for the scenario divided by
the area of the analysis unit) experiences an increment above 10%.
Following the larger administrative divisions used by the Swedish
Forest Agency, regional changes were aggregated into the Swedish
regions: Upper Norrland, Lower Norrland, Svealand and Gtaland, and
shown as percentage changes between 2010 and 2100 (i.e. area change
relative to available area in the region) with error bars showing the
variability in the change between maximum and minimum values
generated among the three climate models. Finally, ranges of service
provision dened by maximum and minimum values among climate
models were plotted for each scenario.
182
Ecosystem Services 23 (2017) 174208
3. Results
3.1. Ecosystem service provision
Timber provision grows steadily for all forest types except nemoral
broadleaf during the rst third of the simulation period across
scenarios, as standing timber stocks are harvested (Fig. 3). Harvests
fall thereafter and grow again during the last third of the century for
the reference scenario, while for the SSP-RCP scenarios they remain
largely unchanged during this period. Nemoral broadleaf timber
provision, however, grows only modestly for all scenarios throughout
the simulation period, while remaining far below demand levels.
Carbon sequestration for the reference scenario mainly decreases
during the rst half of the century and increases during the second
half. For all other scenarios it decreases throughout, though more
slowly during the second half of the century.
Biodiversity and recreation provision generally increase for ap-
proximately the rst 20 years, decrease until the middle of the century
and increase thereafter. While scenarios with greater challenges to
climate change mitigation (i.e. SSPs 3 and 5) tend to cluster together,
others show higher supply levels and more dierentiated trajectories.
Biodiversity demands are met under four scenarios including the
V. Blanco et al.
Fig. 3. (continued)
reference, while demand for recreation is only met under RCP3-
SSP8.5.
Cereal provision grows and in some cases starts to stabilise in the
second half of the simulation, although under no scenario does it reach
demand. Meat supply does however meet demand throughout the
simulation.
3.2. Land-use change
Overall, major land-use change was largely concentrated in the
northernmost region (i.e. Upper Norrland), while the southernmost
(i.e. Gtaland) experienced the smallest changes (Fig. 4a). Changes in
SSP3-RCP4.5, SSP3-RCP8.5 and SSP5-RCP8.5 are largely similar,
although uncertainty in the magnitude of change is larger under
RCP8.5 in the southern half of the country due to more divergent
underlying climate change projections.
Nationally, we nd an increase in agriculture and a decrease in
unmanaged land, with the most extensive use of previously unmanaged
land in the reference and SSP1-RCP4.5 scenarios. For other scenarios,
extensive conversion of unmanaged land concentrates in Norrland.
Agricultural expansion is generally larger under scenarios with higher
mitigation challenges, and tends to concentrate throughout scenarios
in the middle of the country (i.e. Lower Norrland and Svealand).
Monocultural conifer plantations tend to show minimal or no
change. Pine-spruce forests generally expand extensively in the north
and decrease in the south. Under the reference scenario however, they
expand throughout the country. Mixed conifer-broadleaf forests ex-
pand in Upper Norrland and Svealand, but have only small increases in
Lower Norrland. Trends in change for these forests dier in Gtaland,
where expansion happens under the reference scenario and SSP1-
RCP4.5, but no change is observed under SSP4-RCP4.5, and loss
occurs under all other scenarios. Boreal broadleaf forests have no
observable change. Nemoral broadleaf forests undergo either no
change or expansion, the latter being most prominent under SSP1-
RCP4.5.
Hotspots of land-use expansion in pine-spruce forest, mixed
183
Ecosystem Services 23 (2017) 174208
conifer-broadleaf forest, nemoral broadleaf forest, agriculture and
unmanaged land are observable for the dierent scenarios (Fig. 5a).
Western Norrland is a hotspot for pine-spruce forestry expansion
across all scenarios. Pine-spruce forest hotspots occur also in the south
under the reference scenario, and to a lesser extent under SSP4-
RCP4.5, SSP3-RCP4.5, SSP3-RCP8.5 and SSP5-RCP8.5. Only one
hotspot occurs of mixed conifer-broadleaf forest expansion in the
south under SSP1-RCP4.5. This scenario also has several hotspots of
nemoral broadleaf forests in Upper Norrland, Svealand and Gtaland.
Agricultural expansion hotspots are present in the southern regions
under SSP1-RCP4.5, SSP3-RCP4.5 and SSP3-RCP8.5, while under
SSP5-RCP8.5 they also occur towards the north. Finally, a hotspot of
land abandonment exists consistently in south-eastern Gtaland under
all scenarios except the reference.
3.3. Changes in land owner management roles
Upper Norrland was the only region where no land owner manage-
ment role had a discernible decrease across scenarios (Fig. 4b). In all
other regions the percentage of productionists decreased, except in
Gtaland under the reference scenario. Productionist loss was stron-
gest in Svealand and weakest in Gtaland, reecting a decrease in the
proportion of productionists nationally under all scenarios except the
reference. Multi-objective forest owners were most successful, sub-
stantially increasing in numbers except in the south under scenarios
with greater challenges to climate change mitigation. This type
experienced its largest increases in Upper Norrland.
Recreationalists decreased in the southern regions and increased in
Upper Norrland, but their total numbers decreased under all scenarios
except SSP1-RCP4.5 and the reference scenario. Conservationists
experienced small or no changes in all regions and under all scenarios
except for SSP1-RCP4.5, under which their numbers increased sub-
stantially, especially in Upper Norrland. The percentage of passive
owners remained nearly unchanged under all SSP-RCP scenarios, but
increased slightly under the reference scenario, mostly in Gtaland.
Commercial farmers generally increased in numbers under all SSP-
V. Blanco et al.
Fig. 4. Percentage change of a) land-use and b) land owner management role categories per scenario and region. Error bars show the ranges of change generated across the three
climate models for each scenario.
RCP scenarios, but decreased in southern regions under the reference
scenario. Non-commercial farmers show very similar changes under all
scenarios, increasing in the north, barely changing in the mid latitude
regions, and decreasing in the south.
We observe hotspots of increase in productionist, multi-objective,
passive, commercial farmer, and non-commercial farmer management
roles among all scenarios (Fig. 5b). The reference scenario resulted in
hotspots of productionists in Upper Norrland, multi-objective owners
in all regions, and passive owners in Lower Norrland and Gtaland.
Under SSP1-RCP4.5 and SSP4-RCP4.5 multi-objective owner hotspots
occurred in all regions although to a much lesser extent in the southern
ones, where commercial farmer hotspots also occur under SSP1-
RCP4.5. SSP3-RCP4.5 had incremental hotspots in multi-objective
owners in southern regions, and commercial farmers in all regions.
SSP3-RCP8.5 diers from SSP3-RCP4.5 in that it has a hotspot of
multi-objective owners in the south, and it has no commercial farmer
hotspots in Upper Norrland. Finally, SSP5-RCP8.5 has hotspots of
multi-objective owners in all regions, primarily northern ones, com-
mercial farmers throughout the country, and non-commercial farmers
in Upper Norrland.
184
Ecosystem Services 23 (2017) 174208
4. Discussion
4.1. Future changes in land use and ecosystem service provision
This work demonstrates the adoption of mechanisms to simulate
forest owner decision-making in large scale land-use change and ES
provision modelling, and uses these to assess possible changes in the
Swedish forestry sector under dierent climate and socio-economic
change scenarios. Findings suggest substantial scope for model appli-
cations of this kind, and also for great variation in ES provision in the
Swedish forestry sector, and important land-use change throughout the
country.
Some of the changes in ES provision that we simulate are not
caused by scenario conditions or forest manager decision-making. In
particular, the sinusoidal trajectory of pine, spruce and boreal broad-
leaf timber supply through time under the reference scenario is largely
determined by the uneven distribution of forest ages within Sweden.
This is a legacy of past land management decisions. Forestation in the
country increased from the beginning of the 20th century until the
1960s, when highest rates of forest regeneration (i.e. including plant-
V. Blanco et al.
Fig. 4. (continued)
ing, seeding, and natural regeneration) per year were attained, after
which rates began to decline (Swedish Forest Agency, 2015). The
model suggests that this will result in a peak in supply during the
2030s, approximately 70 years after the forest planting peak. In reality,
the magnitude of this peak is likely to depend upon the extent to which
felling can be coordinated to preserve supply and price levels, but the
need to harvest within certain time periods does constrain the scope for
such actions. As a result, decreases are likely in timber revenue and in
the supply of other forest services, with reductions in the supply of
biodiversity, recreation, and carbon sequestration being projected in
our simulations. Therefore, it is clearly important that the impacts of a
future acceleration in timber harvesting are considered and addressed.
More generally, the substantial decoupling between the simulated
supply of, and demand for, forest services over the course of 90 years
(in contrast to the results for agriculture) demonstrates the diculty of
continuously meeting societal demands for ES produced over long
periods of time. The consistent supply of multiple forest services
represents a complex optimisation problem that can only be solved,
if at all, using national overviews and top-down (e.g. policy) mechan-
185
Ecosystem Services 23 (2017) 174208
isms. An obvious example is the provision of biodiversity and recrea-
tion, which depends upon nationally designated areas and prohibitions
on felling in non-productive forests (Swedish Forest Agency, 2014),
even in the most environmentally-friendly scenarios we modelled (i.e.
SSP1-RCP4.5). Future analyses of service provision under such scenar-
ios should therefore also consider services provided by protected
natural systems, in order to evaluate how far they are able to absorb
impacts of changes such as large-scale felling in productive forests.
Another consideration is the extent to which food demand exerts
pressure on forest management and ES provision, with our simulations
suggesting shortfalls in cereal supply even under sub-optimal forest ES
provision. A higher societal sensitivity to cereal supply levels would
contribute to bringing supply closer to the demand, which would
further compromise the provision of forest ES. Furthermore, the
potential increase in agriculture in the north would entail a more
widespread competition with forestry throughout the country. Our
simulations suggest that this northward expansion would largely come
at the expense of unmanaged land, comprising wetlands and semi-
natural vegetation, which also supply ES such as water supply, nutrient
V. Blanco et al.
Fig. 5. Locations of hotspots of increase in a) land-use and b) land owner management role categories under all scenarios in Sweden.
retention, food, recreation, or biodiversity (Costanza et al., 1997).
Hence, it is important to understand that meeting future demands for
agricultural and forest services will likely entail trade-os with ES
supplied by other natural systems.
We also nd that, under the SSP-RCP scenarios, carbon sequestra-
tion decreases as timber is increasingly felled throughout the rst third
of the century, but does not increase again as timber felling decreases
after that (in contrast to the reference scenario). Similarly, biodiversity
and recreation decreases are never entirely reversed under the SSP-
RCP scenarios, while they are under the reference scenario. The reason
for this is that forest-to-forest land-use change is substantially more
frequent under the SSP-RCP scenarios than for the reference, meaning
that fewer forests reach their scheduled age of felling under the SSP-
RCP scenarios. The lower mean national forest age reached under SSP-
186
Ecosystem Services 23 (2017) 174208
RCP scenarios also entailed smaller annual timber harvests nationally
as younger forests were being felled, and this keeps carbon sequestra-
tion, biodiversity and recreation at lower levels. These ndings conrm
qualitative results from Roberge et al. (2016), who suggest that the
provision of supporting and cultural ES, and economic outputs from
timber would be negatively impacted by shorter rotations in
Fennoscandian forests.
This phenomenon is explained by changes in demand levels and
climate change, which together prompt forest managers to adapt their
management activities more frequently and dramatically than they
would otherwise do (with the size of this eect depending on the
balance between the costs of early felling, timber prices and potential
prots from alternative management strategies). This eect is consis-
tent with the ongoing consideration of management alternatives (e.g.
V. Blanco et al.
Fig. 5. (continued)
multi-species planting, or introduction of exotic species respectively) as
adaptations to climate change (Felton et al., 2016; Kjaer et al., 2014). If
forest owners do consider felling early when faced with a protable
alternative, the successful uptake of one or more innovations could
trigger continued changes between forest types as simulated here,
which could negatively aect the provision of services that increase
with forest age. Further simulations including the uptake of innova-
tions as a function of networks among forest owners (Satake et al.,
2007) and relevant institutions would be particularly useful here.
Many of our ndings were broadly consistent between SSP3-
RCP4.5 (Regional rivalry and low emissions) and SSP3-RCP8.5
(Regional rivalry and high emissions), which seems to indicate that
the inuence of climatic change on land productivities (being the only
parameter dierent between them) may be less important than that of
socio-economic changes or behavioural dierences. Additionally, the
great resemblance between SSP3-RCP8.5 and SSP5-RCP8.5 (Fossil-
187
Ecosystem Services 23 (2017) 174208
fuelled development and high emissions) suggests that land owner and
societal behavioural factors (e.g. sensitivity to prot levels on the part
of landowners and sensitivity to supply levels on the part of society),
despite substantially dierent service demand trajectories, are key in
determining the course and impact of land use change. Furthermore,
this holds true for the provision of timber and non-timber forest
services.
Climate change, and especially service demand levels do, however,
appear to inuence the geographical distribution of land-use, as
observed when comparing SSP3-RCP4.5, SSP3-RCP8.5, and SSP5-
RCP8.5. We see an expansion in both commercial and non-commercial
agriculture towards the north as a response to the higher demands for
agricultural products and increasing suitability for agriculture in the
north. Additionally, the distinctive expansion of nemoral broadleaf
forests and conservationists across the country, and of recreationalists
in the north under SSP1-RCP4.5, are consequences of the higher value
V. Blanco et al.
and demand being placed by society on biodiversity and recreation.
4.2. Model assumptions and limitations
It can be dicult to identify causes and eects in simulations from
complex system models because of the multiple interactions and
feedbacks inherent within such models. However, we can identify some
relationships between model inputs and outputs from the sensitivity
analysis and the exploration of simulation results, and sources of
uncertainty among model assumptions. Giving-in and giving-up
thresholds represent a range of personal characteristics that control
an agent's responsiveness to demand levels. Here we assign random
distributions to these thresholds, in the absence of empirical data with
which they could be parameterised. Our, and previous (Brown et al.,
2014, 2016; Murray-Rust et al., 2014), sensitivity analyses show that
the eect of random model components on agent behavioural para-
meters has a lower impact on ES provision compared with components
that dier between scenarios (demand and productivity changes, and
mean behavioural values). Land owner giving-up probabilities help to
regulate the rate of owner type and land-use change, with lower giving-
up probabilities resulting in less abrupt yearly changes, especially for
farmers/farmland. Greater randomness between agents means less
rationality in land use change. Further insight into how behavioural
parameters aect simulation results, through extensive sensitivity
analyses, are presented in Brown et al. (2014, 2016) and Murray-
Rust et al. (2014).
The ES demand change scenarios were derived from the SSP-RCP
scenarios, but it is important to acknowledge the diculty in estimat-
ing such demands because of the substantial uncertainty inherent
within these scenario assumptions. It is also important to be cautious
with the results for owner type and land-use change in the north west
of Sweden, dominated by the Scandes mountain range. In this area, the
eect of topography on productivity changes arising from climate
change is dicult to model since productivities were calculated from
climate change data at a coarser spatial resolution (5050 km).
Therefore, the eect of climate change is highly uncertain in north-
western Sweden. Furthermore, the model allocates forest in the
mountainous region under all scenarios, including the reference, to
cells where forest was previously not present. This is mainly due to the
time lag between the moment when forests start being planted in
response to unmet demands and the moment when demands are met.
During this time period, while service supply from young forests is still
low, forests continue to be planted, to the point that they occupy
available areas that may be less productive. This may be unrealistic
depending on the level of return that real world land owners are
willing to accept from forest activities.
Not allowing land ownership to change between forest owner types
before stand maturation means that the eect of management on forest
growth rates and their service provision throughout the life of the forest
is xed when the forest is planted. This would only be a limitation
though if it were common for forest ownership and management to
change during early forest development stages. Additionally, we did not
account for voluntary set-asides for conservation, which make up at
least 5% of the productive forest area (Swedish Environmental
Protection Agency, 2016). In addition, due to certication require-
ments, trees, tree groups and buer zones are left on most felling sites,
also reducing the harvested volume somewhat. Had voluntary set-
asides and certication requirements been accounted for, yearly timber
provision would have been lower and the outcomes of the modelled
competition process slightly dierent. Conversely, as today, the major-
ity of the planting material in Sweden comes from seed orchards, and
these plants are expected to grow considerably faster (1020%)
188
Ecosystem Services 23 (2017) 174208
compared to plants used in past decades, accounting for this eect
would have led to higher standing volumes and timber supply.
Consequently, the eect on timber provision of not including voluntary
set-asides or certication requirements, and the eect of not including
planting material from seed orchards have opposing eects at the
aggregate level.
The trajectory of timber provision will, however, likely be less
markedly sinusoidal than presented here. The forest age distribution
used (from SLU Forest Map) was shown to overestimate the area
covered by middle aged forest and underestimate that of old forest,
even though more forest was still planted around the 1960s.
Consequently, while timber supply may still peak in the 2030s,
harvesting is likely to be distributed more evenly through time than
presented here. Simulation results presented in the Swedish national
forest impact assessment show a relatively steady increase in timber
provision throughout the century, mainly driven by increased produc-
tivity from climate change (Claesson et al., 2015). This higher
regularity in supply resulted mainly from their inclusion of a more
evenly distributed age distribution (from the National Forest
Inventory). Ranges of levels of total timber provision were largely
similar in their assessment and ours, even though we did not take the
eect of tree breeding into account. CRAFTY-Sweden also considers
some drivers that the national forest impact assessment did not (i.e.
demands for ES and human behaviour), leading to dierent dynamics,
especially towards the end of the century, as discussed below.
5. Conclusions
CRAFTY-Sweden brings us a step forward in the understanding and
representation of large scale land-use change and its complexities
under climate change. Our results show that variability in human
behaviour has a substantial role in determining the eects of climatic
forces and societal demands on ES provision and land-use change.
Important changes in land-use and ecosystem service provision can be
expected in Sweden as a result of changing demands and climate,
especially towards the north of the county. Increasing food demands
and increasingly favourable climatic conditions for agriculture would
lead to agricultural expansion, if food imports are not increased. Such
expansion would require trade-os with currently unmanaged land,
such as wetlands, which contribute important ES. Furthermore,
accelerating timber harvesting throughout Sweden due to a nationally
uneven age distribution, and increasing rates of forest land-use change
between forest types may have negative consequences for forest ES
provision. Finally, the challenge of steadily meeting societal demands
for ES produced over long time periods and at large scales would
require top-down mechanisms that use national-scale information to
regulate forestry processes and, to the extent possible, their conse-
quences.
Acknowledgements
We thank Prof. Ben Smith for his valuable comments at dierent
stages of this research. We also thank two anonymous reviewers for
their constructive and insightful comments on a previous version of
this paper. The research of VB was supported by the Mistra-SWECIA
research programme and the University of Edinburgh. FL, GV and ML
were also supported by the Mistra-SWECIA programme. The work of
CB, SH and MR was performed under the project IMPRESSIONS
(Impacts and Risks from High-End Scenarios: Strategies for Innovative
Solutions) funded by the European Union's Seventh Framework
Programme (Grant Agreement No. 603416).
V. Blanco et al. Ecosystem Services 23 (2017) 174208

Appendix A

ODD1 Protocol for CRAFTY-Sweden

Purpose
The Competition for Resources between Agent Functional Types in Sweden (CRAFTY-Sweden) model is designed to model land-use changes in
Sweden, with a focus on forestry. CRAFTY-Sweden applies the model framework, CRAFTY-CoBRA which is an extension of CRAFTY (Murray-Rust
et al., 2014) that allows for dynamic production functions. CRAFTY can be used to investigate the eects of human behaviour on land use
transitions under a range of socio-economic and environmental scenarios. CRAFTY is designed to be exible, capable of handling a large variety of
data and to be applicable across a wide range of empirical or theoretical settings.
CRAFTY-Sweden is founded on ecient and tractable descriptions of individual behaviour and decision-making that takes account of the eects
of climatic and environmental change, and may be adapted to a range of applications and scenarios. It applies exogenous demand levels, which
agents attempt to meet according to behavioural rules and ecosystem service supply. The model considers the adoption of dierent land uses,
variations in the intensity of land uses, diversication into multifunctional land uses, changes in productivity over time, land abandonment, and
competition for available land.
Agents use capitals that are available for the land parcels that they own and supply ecosystem services based on their respective production
function. Capitals may vary across the modelled landscape and over time. Agents of the same functional role may have the same or heterogeneous
weights for the production function and this together with the available capitals determines the supply level for each service (i.e., timber for dierent
forest types, food, biodiversity, carbon sequestration, and recreation) by the agents. The current demand for a particular service and an agents
productivity determine its competitiveness, which in turn, aects the introduction of new agents in the model: the distribution of agents over a
landscape and the introduction of new agents into the system during simulation are determined by an allocation procedure, which is discussed in
section 7. Submodels, Allocation. Institutions may change capital levels and issue land use restrictions (see section 7. Submodels, Institutions).

Entities, state variables, and scales

Spatial units. CRAFTY-Sweden is based on a grid of cells, representing any absolute spatial scale. Each cell has dened levels of a range of
capitals, which describe the availability of particular social, environmental or economic resources. A non-spatial population is assumed to exist and
to generate demands for services. Each cell may be managed by a single land use agent.
Agents. Forest owners and farmers are explicitly represented as agents in CRAFTY-Sweden. Both share a common architecture where agents are
made up of a functional role (FR) characterising function and role in the system, and a collection of properties.
A land use agent is able to leverage the capitals available on a land parcel (represented as a cell) to provide a range of services. Each agent has
a production function as part of its FR, which maps capital levels onto service provision (see section 7. Submodels, Production). An agents
competitiveness, according to a given level of service provision, can be calculated from societal demands, overall supply levels and marginal utility
functions. See Table A1 for a complete list of agent variables.
An agent searching for land can either take over unmanaged (abandoned) cells, or cells on which it can outcompete an existing agent. Between
them, giving-up and giving-in parameters provide a stylised interpretation of factors that make human behaviour deviate from narrowly dened
optimality, such as personal connection to a landscape or way of life, or resistance to change.
The functional roles (FR) that are assigned to agents make up a typology that denes general characteristics of land manager practices.
Searching agents can be prototypes of specic FRs that allow the comparison of productivity, utility and other characteristics of typical agents of
that FR. Finally, individual agents of a given type need not be identical all of the agents characteristics, including production functions, and giving
up/giving in thresholds are drawn from distributions to provide within-type heterogeneity.
We developed a typology of forest and agricultural agents, focusing especially on the former. To dene forest owner types we used as a basis the
forest owner typology developed by (Blanco et al., 2015). Because this typology was based on studies performed at dierent scales and contexts to
those found in Sweden, we performed a validation exercise of the typology using empirical information from 872 Swedish forest owners (Vulturius
et al. in review). A cluster analysis showed that the ve overarching management roles identied by the theoretical typology (productionist, multi-
objective, recreationalist, conservationist and passive) were also clearly discernible in the empirical data. Supplementary materials on this validation
can be found in Appendix B.1.
Within each overarching management role, dierent options for forest management are possible, including the use of dierent types of forest
(dened by species composition). Forest types were assigned to each management role on the basis of existing forest stand compositions (Swedish
Forest Agency, 2015) and potential adaptation measures to climate change that consider species composition, number of thinnings and rotation
lengths (Felton et al., 2016; Jonsson et al., 2015) on the basis of owner objectives (Blanco et al., 2015; Duncker et al., 2012a). Assigned forest types
were pine (Pinus sylvestris), spruce (Picea abies), boreal broadleaf (Betula pendula, B. pubescens, Alnus incana, A. glutinosa, Populus tremula),
nemoral broadleaf (Fagus sylvatica, Quercus robur, Fraxinus excelsior, Ulmus glabra, Tilia cordata, Carpinus betula), and combinations of these,
resulting in 17 forest owner types.
Given the current levels of agricultural production (Swedish Board of Agriculture, 2009) and management intensities prevailing in Sweden
(Institute of Environmental Studies, 2015), farmers were separated by the main services provided (i.e. cereal or meat) in combination with their
main objectives (i.e. commercial or non-commercial).
Environment. The environment represents the terrain of Sweden with its varying geophysical features. This heterogeneity in the modelled
landscape is represented by the amount of capitals (such as economic, nature, infrastructure) that exists on a cell. For instance, a cell in a forest
region will have much higher natural capital than infrastructural capital.
Scales. A time step in CRAFTY-Sweden represents a year in practice since this is the time span for which land managers make decisions, which is
true at least for farmers who grow perennial crops. Space is represented by a grid of 1 km2 cells.

1
The model description follows the ODD (Overview, Design concepts, Details) protocol for describing individual- and agent-based models (Grimm et al. 2006; 2010).

189
V. Blanco et al. Ecosystem Services 23 (2017) 174208

Table A1
Variables and States of agents and their FR. Default states are given in parenthesis if applicable.

Variable Description States

Competitiveness Denotes the agents current competitiveness value ]-,[

Giving-in threshold During competition, if a competing agents competitiveness is greater than the incumbent agents by a ]-,[
value larger than the giving-in threshold then the incumbent agent relinquishes that cell to the competitor.
Giving-up threshold If an agents competitiveness falls below its giving-up threshold (defines the minimum return an agent is ]-,[
willing to accept from a cell) it needs to abandon the particular cell (considering giving-up probability).
Giving-up probability Probability for giving up in case the agents competitiveness falls below the giving-up threshold [0,1] (1.0)
Functional component
Role Refers to the Functional Role (FR) Reference
Optimal production Amount of produced service in case of optimal conditions (all relevant capitals = 1.0) [0, [ or formula (based on JEPa)
Capital sensitivities Sensitivities of production towards capital values [0,1] (1.0)
Production model Component responsible for the calculation of service provision Simple Production Model or Dynamic Max
Production Model

a
Java Expression Parser (http://www.cse.msu.edu/SENS/Software/jep-2.23/doc/website/index.html).

Process overview and scheduling

At each modelled time step, the level of service production achieved by an agent is given a benet value via a benet function that relates
production levels to unmet demand. Agents compete for land based on these benet values, and this competition is aected by individual or
typological behaviour. Table A2 gives an overview of the CRAFTY-Sweden simulation schedule.
Each time step starts by updating the decision-making context for land use agents the levels of demand, capitals and any active policies. This
has two stages:

Updates are made to the levels of demand across each region, and levels of capitals within each cell. These are typically loaded from external les,
either as direct values or as functions to be sampled from on a yearly basis. Mechanisms are also available to dynamically modify capitals, for
example in order to model land degradation through intensive agriculture, allowing for feedback loops in this SES.
Next, the land use agents respond and adapt to this altered context:
First, each agent updates its level of supply, based on current capital levels. The total supply of each service is then calculated.
Next, each agents competitiveness is calculated, including the eect of any institutional policies.
Any agents who give up are removed from the model.
The active allocation procedure now runs, allowing new agents to take over unmanaged land and allowing other land transitions to take place,
subject to restrictions for certain transitions.

Once all of the land use agents have been updated, nal accounting is carried out, such as calculating total supply and demand, creating output
les, displaying model state and creating model run animations.

Table A2
Simulation schedule.

For each agent agents

increase age
For each agent agents
update competitiveness based on demand_residual
If competitiveness < threshold_giving-up
leave cell
For each region regions
allocate land: for each unmanaged cell
allocate new agent of most competitive functional role
compete for land: for each fr functional_roles
calculate frs competitiveness on perfect cell
For n search iterations
select fr randomly according to competitiveness
For m cells
calculate frs competitiveness
If frs competitiveness > owners threshold_giving-in
owner relinquishes cell
agent with fr takes over cell
For each agent agents
update supply of services produced
For each region regions
update supply and demand_residual
For each agent agents
update competitiveness based on demand_residual
generate output

190
V. Blanco et al. Ecosystem Services 23 (2017) 174208

Design concepts
Basic principles. The design criteria used for the specication of the model framework were:

1) The model must be able to run at large scales. This requirement holds for runtime costs, complexity, and the availability of data to parameterise
and calibrate the model.
2) The model should take into account the full range of societal demands, including those that are not dened explicitly in monetary terms such as
biodiversity.
3) The model must be able to represent multifunctional land use, and be responsive to the trade-os between the provision of various services.
4) The model should be able to represent the diversity of human behaviour and land management.
5) The model must be able to deal with the long-term allocation of forest types.

Agent Functional Types are derived from the concept of Plant Functional Types in Dynamic Vegetation Models (e.g. Lavorel et al., 2007) and
used to group land-use agents by their decision making and productive behaviour; here adopted as Functional Roles. Land-use modellers are
familiar with the use of typologies, especially in constructing agent-based models as representations of real-world actors (Robinson et al., 2007;
Valbuena et al., 2008). Typologies allow generalisations of the attributes (traits) of individual actors in a system that simplies model development
and application, and provides a more transparent representation of agent decisional processes and behaviour.
Table A3 provides an overview of the assumptions that guided the model framework development.
Emergence. Emergent eects that could be observed as outcomes of experiments using CRAFTY-Sweden are spatially explicit changes to land
ownership and management, the intensication of land uses, including mono- or multi-functional land uses, changes in productivities and yields of
dierent land uses, and eects on capital levels.
Adaptation. Land use agents may change their functional role in response to unsatisfactory competitiveness, e.g., due to changes in demands for
the service they produced so far.
Objectives. The land managers objective is to be suciently competitive in the supply of societal benets.
Learning.There is no learning in CRAFTY-Sweden.
Fitness. Agents survival in the system depends upon their competitiveness, which is determined by an agents ability to meet the demand of
services in a modelled society (see Section 7, Population, Services, Demand and Utility).
Prediction. Agents in CRAFTY-Sweden do not explicitly predict.
Sensing. Land use agents in CRAFTY-Sweden are aware of the current demand (regional or global depending upon the setup) that is to be met.
They consider the production they are able to achieve on a given cell with a particular capital level, e.g., when they want to change their functional
role. Agents are aware of the competitiveness of other agents and may relinquish their cells to agents that are more competitive.
Interaction. Direct interactions occur between new (potential) and existing agents that compete for cell ownership. Land owner agents compete
for land on the basis of their benet values, which depend upon their ability to produce services and societal demand levels for those services. As
land ownership and management change, demand and supply levels also change, so that actions taken by each agent aect the decisions of others.
Stochasticity. There are a number of stochastic processes in CRAFTY-Sweden:

Initialisation of agent properties (e.g., giving-in/up thresholds, optimal production, capital sensitivities) from probabilistic distributions and/or
the addition of noise
Search for cells during competition
Selection of functional roles to compete with incumbent land managers
Probabilistic giving-up
Forest stand felling age

Random numbers are sourced from dierent random streams which allows their separate control via dened initial random seeds (see CRAFTY
online documentation2). This also ensures the reproducibility of simulations results.
Collectives. Two types of agent collectives exist during a course of a simulation run. First is the list of agents that possess land parcels (cells) in
the simulated landscape (grid). Second is the list of potential agents that enter the system to takeover cells from existing agents (if possible) or
occupy a vacant or abandoned cell on the grid.
Observation. CRAFTY provides a range of observations and displays to help understand the model behaviour during runtime. Each of the
submodels has a display, which is either numeric or graphical, showing curves for variables of note. A range of spatially explicit outputs is also
available; these include agent type, capital levels, competitiveness scores, supply of services, and so on. Any of these displays can be used to create
videos of the models behaviour over time.
Furthermore, CRAFTY enables the output of a number of simulation data as CSV or raster les.

Initialisation

To allow the frameworks conguration by non-programmers it is accomplished by a set of interlinked XML and CSV les. XML les dene basic
simulation parameters and provide properties for the initialisation of model components coded as Java objects, while CSV les provide data when
there are many values required. The approach is highly exible and extendable.
CRAFTY-Sweden is initialised by reading the le Scenario.xml and following the links therein to the conguration of outputs and the world
conguration, which in turn contains links to model components such as functional roles, the competition model, or the allocation model. A le
Cells.csv includes the coordinates and capital levels of the cells in a region, the initial allocation of agents on these cells, and agents properties such as
functional role, which are applied when these agents are initialised. Figure 1 gives an overview of a possible setting of XML and CSV les (Fig. A1).

2
https://www.wiki.ed.ac.uk/display/CRAFTY/Randomisation

191
V. Blanco et al. Ecosystem Services 23 (2017) 174208

Table A3
Design assumptions made in CRAFTY-Sweden.

Model assumption Details Justication

A wide range of land-use relevant behaviour can
be represented by giving-in and giving-up
thresholds.
Each cell is managed by a single agent.
Land managers can be Functional Roles (FR).
Potential productivity of land can be represented
by a range of capitals.
Production of services by land managers can be
described by a function dependent upon
access to capitals and productive abilities.
The competitiveness of land managers depends
upon demand for specic services.
Three mechanisms of land use change.
Lock forest management approach until the forest
has reached maturity.
A forest owner will not fell a forest before a
minimum felling age.
Passive owners do not plant their forest, and their
production is dependent on the management
of the agent they take over the forest from.
Range of personal characteristics and behaviours known to
affect land use decisions can be often abstracted in two making (e.g. Siebert et al., 2006; Gorton et al., 2008;
values giving (relative) willingness of land managers to
change land use or abandon land. Believed to be a necessary
simplification for large-scale land use models that
adequately mimics observed behaviour but can be
overwritten by more specific decisions (see Sections 2.
Entities, state variables, and scales, Agents and 7.
Submodels, Decision Making).
Multiple ownership of cells is not supported.
The management practices (FR) and behaviour (BT) of land
managers allows them to be classified into a typology
analogous to the Plant Functional Types used in Dynamic
Global Vegetation Models, increasing modelling efficiency.
Capitals representing natural productivity (for any good or
service such as a specific food or timber) and any
anthropogenic effects on productivity (such as availability of
finance or infrastructure) can be used as a basis for the
description of ecosystem services.
The ability of land managers to produce services is
dependent on the underlying productivity of the land,
expressed via capitals (above) and their individual or
typological productive ability, which may depend upon a
number of personal characteristics and behavioural factors.
(A Cobb-Douglas function is used, adapted to incorporate a
time component for forest services.)
Demands exist for ecosystem goods and services, and land
managers compete to satisfy these demands. Land managers
are more competitive when they can produce greater (total)
quantities of services for which there is unmet demand.
Land use (or ownership) changes when agents abandon land,
take over unmanaged land, or take over managed land from
the current owner.
Forest owners will not abandon or change the management
approach on their land until the forest has reached maturity,
in order to recover the initial investment.
A forest owner will not fell a forest before a minimum felling
age dependent on site quality (i.e. productivities), which is
determined by law in Sweden for pine and spruce, and for
which recommendations exist for other species.
They therefore only take over the forest and associated
optimal production function of other owner types managing
forests with the same tree species as them, but do decide
about the forest age at felling.
Known that numerous factors affect personal decision-
Valbuena et al. 2010; Meyfroidt, 2012) - too many to model
or parameterise. Several studies have suggested that, for
modelling purposes, a wide range of behaviours are reducible
to a small number of dimensions similar to those used here
(e.g. Berger, 2001; Polhill et al., 2001; Siebert et al., 2006;
Gorton et al., 2008; Murray-Rust et al., 2011).
The scale of application is not defined and so can be set to
the appropriate scale of land holdings in any particular case
(the minimum size of holding that is of interest to the
modeller). Agents may be permitted to manage multiple
cells.
The use of types increases computational efficiency by
providing a description of land management and human
behaviour at a level of abstraction that decreases the need for
empirical parameterisation but retains the characteristics
most important to large-scale land use change (Arneth et al.,
2014). Splitting in FR and BT allows changes in one
component while the other persists.
Well-established method of characterising land see
Boumans et al. (2002) and Scoones (2014).
Douglas (1976), Fulginiti and Perrin (1998) and Martin and
Mitra (2001)
Demands for services are known to be expressed via the
economic value of service production, and, in the absence of
behavioural factors, land use is driven primarily by
economics. Partly, decisions are made on grounds of non-
monetary (or indirectly monetary) demands e.g. for green
space, fresh water etc. and CRAFTY-Sweden is designed to
be capable of handling these, where they can be
parameterised. No assumption about the relationship
between unmet (residual) demand and utility values
(competitiveness) is made.
The same limited number of options are possible in the real
world.
Based on expert knowledge from Swedish forestry
researchers
Forests are felled in Sweden after reaching an age that
depends on site quality (Lagergren et al. 2012). The stand
age at felling is regulated in law for pine and spruce to
guarantee that the production potential is utilised (Kunskap
and Direkt, 2015), and for beech, birch and oak
recommended rotation periods exist (Lf et al., 2009; Rytter
et al., 2008).
Passive owners generalised lack of primary objectives for
forestry (Blanco et al., 2015).

Input data
Land cover data. To create a baseline land ownership map for 2010 we rst devised a land-use map at 1 km2 resolution that included pine,
spruce, pine-spruce, pine-boreal broadleaf, spruce-boreal broadleaf, boreal broadleaf, and nemoral broadleaf productive forests, agriculture,
protected areas, non-productive forests, semi-natural vegetation, wetlands, open spaces, other unmanaged land, articial, and water bodies. SLU
Forest Map data (SLU, 2010) on the proportion of dierent tree species per cell were used to identify forest cover and classify it according to the
forest types assigned above, according to the proportion of forest within the cell, and the proportions of dierent species within that forest. CORINE
land cover (EEA, 2014) was used to identify all other land use/land cover classes. Nationally Designated Areas (EEA, 2015) were then superimposed
to dene protected areas. Non-productive forests are also protected and unavailable for production (Swedish Forest Agency, 2014). Thus, we
identied them by:

192
V. Blanco et al. Ecosystem Services 23 (2017) 174208

Fig. A1. Overview of model conguration, showing relationships between les and what each le provides.

Table A4
Identities and data sources for modelled capitals, and the ecosystem services they contribute to producing.

Capital Denition Input data Ecosystem services Data source

(units; resolution)

Pine, Spruce, Boreal Broadleaf, and
Nemoral Broadleaf Forest
Productivities
Grassland Productivity
Transportation Infrastructure
Baseline productive potential
for each forest type
Baseline productive potential
for grassland and cropland
Proximity to transportation
networks and central markets
Forest production potential per forest type 1. Pine, spruce, boreal br. (Hgglund and
and nemoral br. timber Lundmark, 1987)
2. Carbon (Johansson et al., 2013)
(m3sk ha1 yr1; 1 km2) 3. Biodiversity SLU
LPJ-GUESS simulated C3-grass NPP 1. Meat Simulations done for this
projection for 2010 driven by climate 2. Cereal article (see Section 2.2)
(radiation, temperature, precipitation)
(kg C m2 yr1; 5050 km)
1. Road and rail networks 1. Pine, spruce, boreal br. 1. UNECE
2. Waterway networks and nemoral br. timber 2. EEA
3. Travel time to nearest town with over 50000 2. Meat 3. GEMU, JRC
inhabitants (1 km2) 3. Cereal
4. Recreation

SLU: SLU Forest Map, produced by Swedish University of Agricultural Sciences, accessed via ftp://salix.slu.se/download/skogskarta
UNECE: United Nations Economic Commission for Europe, accessed via http://www.unece.org/trans/areas-of-work/transport-statistics/statistics-and-data-online.html
EEA: European Environment Agency, accessed via http://www.eea.europa.eu/data-and-maps
GEMU, JRC: Global Environment Monitoring Unit, managed by the Joint Research Centre, accessed via http://www.edenextdata.com/?q=content/jrc-accessibility-map-estimated-
travel-time-nearest-city-population-50000

3. Assigning to forested cells the value of the highest productivity found among all forest types within that cell; and
4. Given the proportion of non-productive forest per county (Swedish Forest Agency, 2015), selecting for each county the equivalent number of cells
with the lowest productivity values.

Mean forest age values from the SLU Forest Map were used to assign forest ages.
Forest owner types were allocated to productive forest types using data on a) the area of productive forest land by county and ownership classes
for 2010 (Swedish Forest Agency, 2015); and b) the proportion of owners in each county belonging to each group from the cluster analysis.
Agricultural land and (some) semi-natural vegetation were assigned to commercial cereal, non-commercial cereal, commercial livestock, and non-
commercial livestock farmer agents according to the land-use intensity in 2010 (Institute of Environmental Studies, 2015). The remaining semi-
natural vegetation, wetlands, and other unmanaged land were left unallocated. Protected areas, non-productive forests, open spaces with little or
no vegetation, articial surfaces, and water bodies were made unavailable for allocation during simulations. Fig. B2 shows the resulting map of
Sweden. Further detail on the creation of land-use and land owner type maps can be found in Appendix B.3.
Capital levels. The capitals that agents can use in service production are productivities for pine, spruce, boreal broadleaf, and nemoral broadleaf
forests, grassland productivity (natural capital), and transportation infrastructure (infrastructure capital). Table A4 gives capital descriptions, their
data sources, and the ecosystem services they contribute to producing.
Land productivity levels can be aected by climate change. The ecosystem model LPJ-GUESS (Smith et al., 2001) was used to simulate forest
dynamics during 20102100 using climate projections of the Global Circulation Model-Regional Circulation Model ensembles (hereupon climate
models) EC-Earth-RCA4, IPSL-RCA4 and NorESM-RCA4 for RCPs 4.5 and 8.5 from the EURO-CORDEX project (Jacob et al., 2014; Jones et al.,
2011). Annual climate-induced change was calculated for all productivities using LPJ-GUESS spatial projections of yearly timber volume growth for
pine, spruce, boreal broadleaf and nemoral broadleaf forests, and yearly net primary productivity (NPP) change for grass until 2100 at 5050 km
resolution. After checking for non-linearities in the volume growth and NPP change projections, linear models were considered to be sucient to
represent them. Thus, a regression coecient was calculated for every cell by performing linear regression on projected growth values. These values
were then downscaled to 1 km2. See Appendix B.6 for more details about the calculation of climate impacts on productivities.
Demand levels. Demand levels (for ecosystem services) are exogenous to the model and are dened prior to model initialisation based on land-
use and an interpretation for Europe of the storylines of the Shared Socio-economic pathways (SSPs) (Carter et al., 2015) from Engstrom et al

193
V. Blanco et al. Ecosystem Services 23 (2017) 174208

(2016) and Kok et al. (2015) respectively. Quantications of demands were done for SSP1, SSP3, SSP4, and SSP5. Baseline demands for timber,
cereal and meat were assumed to be equal to observed production in 2010 (FAO, 2015; Swedish Forest Agency, 2015), while those for carbon
sequestration, biodiversity and recreation were assumed equal to simulated baseline supply due to lack of empirical data.
Future projections were calculated using the IIASA SSP data (IIASA, 2015) on decadal rates of change of global forest land cover (for timber and
carbon sequestration), and crop and livestock demands. Demands for biodiversity were projected following the SSP storylines and with guidance
from modelled global future changes in species abundance from UNEP (2007). Rates of change in recreation demands were assumed to be the same
as those for biodiversity.

Submodels
Allocation Model. Land ownership within CRAFTY-Sweden changes according to three dierent mechanisms, which simulate both individual
and collective aspects of land use dynamics. Firstly, agents may abandon their land owing to the competitiveness score that falls below an agents
giving-up threshold.
Secondly, when land is unmanaged, due to abandonment or lack of managers, it can be taken over by a newly created agent. By default, the set of
functional roles is evaluated to determine agent competitiveness score on each unmanaged cell (ca, i). The functional roles are sampled such that the
probability of a role a attempting to take over a cell scales with its competitiveness on a cell with perfect capital levels; P (a ) ca, i , where =0 gives
a random selection and tends towards optimal selection.
For more general land use transitions, an allocation procedure runs between active and potential or ambulant agents to determine ownership
changes. This can include direct competition, where incoming agents attempt to take over existing cells; such an attempt succeeds where the new
agent has a competitiveness on the cell greater than or equal to the existing agents competitiveness plus its giving-in threshold:
cnewccurr +giving_incurr .
Production function. The production of agricultural services is modelled on a yearly basis. Forestry services are however dependent on forest
age. Additionally, climatic change can aect service production by acting on productivities. We developed therefore ways of modelling the time-
dependent component of the dierent services in CRAFTY. The production of a service by an agent in a given year is based on the Cobb Douglas
function, adapted to incorporate a time component (Eq. (1)).

ps = os, t c (ci +ci,t )c (1)

Production ( ps ) of a service s within a cell is the product for all capitals (c ) of the optimal production that an agent type would be able to achieve
in a given year (ost ), and the unit-less (i.e. [0,1]) cell capitals (ci ) plus annual climate-induced change in cell capitals (cit ), weighted by the capital
sensitivities of that agent type (c ) (Table A5). To reect individual variability, optimal production ost is uniformly randomly drawn from
[0.95os, t ,1.05os, t ], and capital sensitivity levels from [c 0.1,c +0.1]. Production calculations for each service are described below.

Timber
For timber production, ost is given by a forest owner type-specic function that determines timber growth given forest age. The ProdMod model
(Eko, 1985) was used to generate timber growth curves for each owner type given their management preferences. Given passive owners generalised
lack of primary objectives for forestry, we assumed them to inherit forest land, and therefore only enabled them to take over the forest and
associated optimal production function of other owner types managing forests with the same tree species. Hence, optimal production functions were
not calculated for passive owners. Table A5 shows parameter values used in ProdMod that diered for each owner type. See Appendix B.4.1 for
further detail on optimal timber production function calculation.

Carbon sequestration
Due to the diculty of calculating soil carbon levels in interaction with forest productivities, only above-ground sequestered carbon (excluding
the stump) was calculated. Optimal production functions of above ground carbon were also calculated using ProdMod outputs (Appendix B.4.2).

Biodiversity
The calculation of optimal forest biodiversity production considered forest age (Duncker et al., 2012b; Koskela et al., 2007; Marchetti, 2004),
using the generation of coarse woody debris with age as a proxy (e.g. Berg et al., 1994; Jonsell et al., 1998; Siitonen, 2001), tree diversity (Gamfeldt
et al., 2013; Marchetti, 2004) and management practices undertaken by each owner type (e.g. woody debris removal), which have an inuence on
biodiversity (Blanco et al., 2015; Duncker et al., 2012a, 2012b). We chose these forest attributes as indicators of biodiversity because of the
availability of baseline data and the possibility of updating the data during model simulations. Finally, we considered the eect of forest productivity
on biodiversity, specically on coarse woody debris (Sturtevant et al., 1997), by assigning sensitivities to timber productivities. For further details of
the calculation of optimal biodiversity production functions see Appendix B.4.3.

Recreation
Recreational value in Scandinavia is largely determined by the age of a forest, but also by forest management practices, accessibility and, to a
lesser extent, by the types of tree species present (i.e. conifer vs broadleaf, and monoculture vs mixed) (Edwards et al., 2012). See Appendix B.4.4 for
further detail on optimal recreation function calculation.

Cereal and meat

Given baseline maps with available capitals and commercial cereal, non-commercial cereal, commercial livestock and non-commercial livestock
agent locations (see Section 2.1.6), their os and c were adjusted until total cereal and meat production equalled the total production in Sweden
reported by the FAO (2015) for 2010. The production of non-commercial agents was set at 0.6 times that of the commercial agents to reect
approximate dierences in production potentials across equivalent classes in Van Asselen and Verburg (2013a, 2013b).
Timber within a cell is harvested and all service provision is set to zero when a forest is clear-felled. The forest in a cell is clear-felled when it
reaches an age that depends on site quality (i.e. productivity) (Lagergren et al., 2012) and owner objectives. In Sweden, the stand age at felling is
regulated in law for pine and spruce to guarantee that the production potential is utilised (Kunskap Direkt, 2015), and for beech, birch and oak

194
 Table A5
Land owner type production, felling age and competitiveness (scenario-independent) parameters. Number of stems planted per ha for each forest type, site index, number of thinnings implemented, age at each thinning per forest type, and
percentage removed per thinning are parameters given to ProdMod to calculate the (age-dependent) optimal timber production and (above ground) carbon sequestration functions. These functions and remaining parameters in this table are
V. Blanco et al.

CRAFTY-Sweden inputs. Yearly os illustrates yearly optimal farmer production. Productivity and infrastructure sensitivities ( ) are given per service. Felling age means ( ) and standard deviations ( ) represent number of years past minimum
felling age (m.f.a.) of a forest, given as the m.f.a. range dependent on site quality.

Service Production Felling Age Competitiveness

Land Owner Type No. Stems/ Site Index dm No. Thinnings Age at each % Removed per Yearlyos m.f.a. , Years Probability
ha Thinning Thinning (tonnes) Productiv. Infrastr. Range past m.f.a. Giving-up

Productionist 2500 280 3 24, 39, 54 25, 20, 20 0.8a, 1e, 0.06f 0.2a, 1g 65100 12, 10 0.05
Pine
b e f b g
Productionist 2600 360 3 23, 38, 53 25, 20, 20 0.8 , 1 , 0.06 0.2 , 1 4595 10, 8 0.05
Spruce
a,b e f a,b g
Productionist 1250, 1300 280, 360 3 24/23, 39/38, 54/53 25, 20, 20 0.8 , 1 , 0.06 0.2 ,1 4595 10, 8 0.05
Pine-Spruce
Productionist 2200 320 3 15, 30, 45 25, 20, 20 0.8c, 1e, 0.06f 0.2c, 1g 4060 9, 7 0.05
Boreal Br.
Multi-objective 1150, 1250 280, 360 2 24/23, 39/38 30, 25 0.85a,b, 1e, 0.06f 0.1a, b, 0.8g 4595 15, 12 0.05
Pine-Spruce
a,c e f a,c g
Multi-objective 1840, 420 280, 320 2 24/20, 39/35 25/50, 20/25 0.85 , 1 , 0.06 0.1 , 0.8 65100 10, 8 0.05
Pine-Boreal Br.
Multi-objective 2000, 420 360, 320 2 23/20, 38/35 25/45, 20/25 0.85b,c, 1e, 0.06f 0.1b,c, 0.8g 4595 10, 8 0.05
Spruce-Boreal Br.
Multi-objective 2100 320 2 15, 30 30, 25 0.85c, 1e, 0.06f 0.1c, 0.8g 4060 15, 12 0.05
Boreal Br.
a,b e f a,b g
Recreationalist 1100, 1100 280, 360 3 24/23, 39/38, 54/53 25, 20, 20 0.9 , 1 , 0.06 0.3 , 0.6 4595 80, 14 0.05

195
Pine-Spruce
c e f c g
Recreationalist 2000 320 3 15, 30, 45 25, 20, 20 0.9 , 1 , 0.06 0.3 , 0.6 4060 100, 14 0.05
Boreal Br.
Recreationalist 1250, 1250 350, 300 3 25/22, 40/37, 55/52 25, 20, 20 0.9d, 1e, 0.06f 0.3d, 0.6g 110150 60, 14 0.05
Nemoral Br.
Conservationist 2100 320 1 15 35 0.9c, 1e, 0.06f 0.3c, 0.8g 4060 100, 14 0.05
Boreal Br.
d e f d g
Conservationist 1250, 1250 350, 300 1 25/22 35 0.9 , 1 , 0.06 0.3 , 0.8 110150 60, 14 0.05
Nemoral Br.
Passive 0.9a,c, 1e, 0.06f 0.1a,c, 1g 65100 25, 17 0.05
Pine-Boreal Br.
Passive 0.9b,c, 1e, 0.06f 0.1b,c, 1g 4595 25, 17 0.05
Spruce-Boreal Br.
c e f c g
Passive 0.9 , 1 , 0.06 0.1 , 1 4060 15, 10 0.05
Boreal Br.
d e f d g
Passive 0.9 , 1 , 0.06 0.1 , 1 110150 10, 10 0.05
Nemoral Br.
Commercial 201 0.8h 0.5h 0.2
Cereal
Non-commercial 121 0.5h 0.3g,h 0.2
Cereal
i i
Commercial 324 0.6 0.5 0.2
Livestock
i
Non-commercial 193 0.3i 0.2g, 0.2
Livestock

a
Pine timber, bspruce timber, cboreal broadleaf timber, dnemoral broadleaf timber, ecarbon sequestration, fbiodiversity, grecreation, hcereal, imeat.
Ecosystem Services 23 (2017) 174208
V. Blanco et al. Ecosystem Services 23 (2017) 174208

recommended rotation periods exist (Lf et al., 2009; Rytter et al., 2008). Hence, lowest minimum felling age was assigned to the highest
productivity values, while highest minimum felling age corresponded to the lowest productivity values (Table A5). Also, each owner type was
assigned a Gaussian distribution of the planned felling age (above minimum felling age) (Table A5). This distribution was dened as being within
the recommended rotation periods for all owner types except for recreationalists, conservationists and passive owners managing broadleaf forests.
As these latter groups are not primarily interested in timber production (Blanco et al., 2015), they were assigned felling age distributions beyond the
recommended rotation period. Felling age is determined at the time that an agent is allocated to a cell by randomly drawing a number (i.e. forest
age) from within the agent types distribution. Upon felling, timber is harvested and carbon that was being sequestered in standing timber is
removed from the national pool.
Population, Services, Demand and Benet. We assume the presence of a population that has a certain level of demand for services D. This
represents the needs of the population for consumables such as food and timber, and less tangible demands such as those for biodiversity or
recreation (excluding those demands which are fullled by imports). The dierence between the supply and the demand of ecosystem services is the
residual (or unmet) demand, R. The marginal benet of production (i.e., the benet attributed to the production of one additional unit of a service)
is a function of this residual demand:
ms=us (rs ); (2)
where ms is the marginal benet for service s, us is a linear function that describes the benet of production of service s and rs is the residual
demand for service s. As u(r) can take negative values, overproduction is actively penalised. For a given bundle of service provision (typically that
provided by an agent leveraging a cell), the competitiveness (or benet) is given by:

US = ps ms;
S (3)

Appendix B

The following document gives supplementary details on the CRAFTY-Sweden model and the methodologies behind its development.

B1 Land owner types - Typology validation

Methods
The validation of the theoretical forest owner typology was done through a comparison with an empirical typology. The empirical typology was
based on survey data from a questionnaire distributed among a randomized sample of 3000 Swedish non-industrial private forest owners, which
had a response rate of 32% (Vulturius et al. in review). The statistical software R (R Development Core Team, 2008) was used to perform a cluster
analysis using Wards method (Ward, 1963), in which 91% (i.e. 872 owners) of the respondents (i.e. those who answered all questions used in the
cluster analysis) were included.
Wards method is an agglomerative hierarchal cluster analysis. Hierarchal clustering is an attempt to optimize, stepwise a subdivision (divisive)
or synthesis (agglomerative) of data. In the case of agglomerative hierarchal clustering, all data points (n) are initially viewed as individual clusters
and the stepwise optimisation clusters the data into one group containing all points. At each step, individuals or clusters of individuals are merged
together with the most similar individual or group.
A new data frame was created containing only those variables to be considered in the cluster analysis. These were the most similar variables,
available from the survey, to those that were found to characterize forest owners in the theoretical supranational typology. These variables included
a number of ownership objectives (timber production, tax planning, biofuels, return on investment, income, recreation, berry picking, aesthetic
value, biodiversity, environmental protection, water protection, tradition, hunting), and socio-economic attributes (gender, income dependency,
property size, experience as forest owners, location of residence). A dissimilarity matrix was created, using cluster package, daisy command with
metric gower. The cluster analysis was performed with the Wards method using the "agnes" command.
A Chi-square test was applied to test for independency between clusters and dierent variables. This determined how important the dierent
variables were for the formation of the clusters.
As resulting clusters can be further aggregated or subdivided into sub-clusters depending on the level of aggregation, we looked at two cluster
groups (one of which emerged from the other) with a number of clusters close to the number of owner functional types in the theoretical typology.
The rst group contained four clusters, while the following group contained eight clusters. These clusters were characterized on the basis of agent
objectives and socio-economic attributes. While the attributes, personal education and possession of a forest management plan, were not included
in the cluster analysis, they were included in the characterisation of clusters. Subsequently, clusters were labeled according to their most prominent
traits. This resulted in an empirical typology comparable to the theoretical typology. As the empirical typology was only representative for non-
industrial private forest owners, a comparison was made with this in mind (e.g. industrial productionists from the theoretical typology would not be
included in the comparison).
To have a systematic way of comparing the two typologies, we tabulated the importance of objectives and the socio-economic attributes of the
dierent clusters and sub-clusters in a similar table to the one used for the supranational typology developed by Blanco et al. (2015). We used
cluster analysis results, which show proportions of respondents who valued an objective to a certain degree along a ve point Likert scale (i.e. from
not important to very important), to determine which objectives were primary, or secondary, or neither. If the mean value along the scale (05) was
> 4, we considered the objective to be primary. If the value was 34, the objective was secondary. While the objectives included in the cluster
analysis were in some cases not the same as in the supranational typology, several objectives in the analysis could be grouped under one objective
from the typology. In this way, timber production, tax planning, biofuels, return on investment, and income corresponded to prot-making; berry
picking and recreation corresponded to personal enjoyment; and environmental protection and water protection corresponded to environmental
quality. If one of the sub-objectives (e.g. timber production) in each of these groups was a primary or secondary objective, we considered the

196
 Table B.1
a) Clusters of non-industrial private forest owners, which were compared to b) forest owner functional types from the typology of Blanco et al. (2015). Following the characterisation from Blanco et al. (2015)), owner clusters were featured through
their primary () or secondary () objectives and socio-economic attributes. Income dependency and location of residence (i.e. residents (R) vs absentees (A)) can be Low (L), Medium (M), and High (H). Educational level, forestry knowledge
and property size are categorised for each cluster in relative terms (Lower, Medium (Med.), Higher) with respect to the other clusters. Cluster and owner functional types in bold represent the rst level of a hierarchy that branch into one or two
V. Blanco et al.

consecutive levels respectively, where owner groups not in bold can be found.

a) C1 C1.1 C1.2 C2 C2.1 C2.2 C2.3 C3 C3.1 C3.2 C4

Objectives Profit-making
Private Consumption
Personal Enjoyment
Public Recreation
Aesthetics (3.91)
Nature Conservation (3.95)
Environmental Quality (3.05) (3.88) (3.98)
Cultural Conservation (4.00)
Hunting (3.09) (2.95)
Privacy

Attributes Age
Educational Levela Lower Lower Lower Lower Lower Med. Higher Med. Med. Lower Higher
Forestry Knowledge Higher Med. Lower Higher Higher Med. Higher Lower Lower Med. Lower
Gender (F/M) 0.23 0.01 0.85 0.23 0.00 0.64 0.69 0.25 0.27 0.16 0.96
Income Dependency M L M-H M M L M-H L L L L
Property Size Larger Larger Larger Larger Larger Med. Much Larger Smaller Smaller Smaller Larger
Location of Residence R: H R: H R: H R: H R: H R: H R: M R: H R: H R: H R: L
A: L A: L A: L A: L A: L A: L A: M A: L A: L A: L A: H
Property Acquisition
Forest Mgmt. Plan (% holders)a 58 55 62 72 73 66 81 27 27 26 67

197
b) Prot- Productionist Industrial Non- For-profit Multiobjective For-profit Non-profit Recreationalist Conservationist Species Ecosystem Passive
oriented productionist industrial recreationist multiobjective multiobjective conservationist conservationist
productionist

Objectives Profit-making
Private
Consumption
Personal
Enjoyment
Public
Recreation
Aesthetics
Nature
Conservation
Environmental
Quality
Cultural
Conservation
Hunting
Privacy
Attributes Age
Educational Lower Lower Higher Higher
Level
Forestry Med. Higher Lower
Knowledge
Gender
Income L, M, H L, M, H L, M, H L, M L, M L L, M
Dependency
Property Size Greatly larger Larger Larger Larger Med. Smaller Smaller Smaller
(continued on next page)
Ecosystem Services 23 (2017) 174208
 V. Blanco et al.

Table B.1 (continued)

b) Prot- Productionist Industrial Non- For-profit Multiobjective For-profit Non-profit Recreationalist Conservationist Species Ecosystem Passive
oriented productionist industrial recreationist multiobjective multiobjective conservationist conservationist
productionist

Location of R: H R: H R: L R: L R: L
Residence A: L A: L A: H A: H A: H

Property
Acquisition
Forest Mgmt.
Plan
a
Variable not included in the cluster analysis.

198
Ecosystem Services 23 (2017) 174208
V. Blanco et al. Ecosystem Services 23 (2017) 174208

(broader) objective to be primary or secondary. Socio-economic attributes included in the cluster analysis almost matched one to one the attributes
in the supranational typology with the exception of age and property acquisition, which were absent from the analysis.

Results
Cluster characteristics are given in Table B.1. We observe distinct owner groups according to their objectives and socio-economic attributes both
in the four cluster and eight cluster groups. In the four cluster group we distinguish two clusters (labeled c1 and c2) with prot-making as their
primary objective, and with several other secondary objectives. These resemble substantially the non-industrial productionists and multi-objective
owners found in the supranational typology. By contrast, the two other clusters (labeled c3 and c4) show low interest in prot-making, while other
objectives are important such as recreational and environmental aspects. In the absence of very important objectives, these clusters are both to a
large extent comparable with recreationalists, conservationists and passive owners from the supranational typology.
Cluster c1 was subdivided into clusters c1.1 and c1.2. While the objectives of c1.1 seem to align with those of a multi-objective owner type, its
socio-demographic attributes align closely with those of multi-objective and productionist owners. Cluster c1.2 however, having a broad range of
objectives with a focus on prot-making and recreational aspects, resembles the multi-objective type.
Cluster c2 breaks down into clusters c2.1, c2.2, and c2.3. Cluster c2.1 aligns very closely with both the non-industrial productionist and the for-
prot multi-objective type, while c2.2 resembles the non-prot, multi-objective and the conservationist types. C2.3, with several primary objectives,
aligns well with the multi-objective owner type.
Cluster c3 was subdivided into clusters c3.1 and c3.2. While c3.1 aligns well with the recreationalist type, c3.2, with owners who give little value
to all objectives, it has the characteristics of the passive owner type. Finally, cluster c4, which did not subdivide (at the same level as with the other
clusters), resembles recreationalists the most, even though it does not have personal enjoyment as a primary objective.

Discussion
Forest owner types in Sweden resemble closely those observed at the supranational level. Most owner types identied in the theoretical typology
could be identied among the clusters of non-industrial private forest owners. As industrial and public owners were not surveyed, their empirical
categorization was not carried out, and they were therefore not included in the empirical typology. While validation for the Swedish forestry system
could not be done using the same method, it is axiomatic within Sweden that the objectives of industrial productionists are for timber production
(i.e. primarily economically orientated). Additionally, because Swedish public authorities in forestry matters support a balance between
environmental and economic objectives, we assumed in the simulation that public owners correspond to the multi-objective owner type.
Hence, we validated a supra-national theoretical forest owner typology by developing an empirical forest owner typology using survey data
relevant to the study area, and comparing them. We can conrm therefore that the theoretical typology is applicable as baseline information in the
parameterization and simulation of forest owner decision making in the land-use context.

B.2 Capitals
Capitals that agents use in service production are productivities for pine, spruce, boreal broadleaf, and nemoral broadleaf forests, grassland
productivity (for cereal and meat production), and transportation infrastructure. Baseline productivities for the dierent forest types reect site-
specic dierences in production potential, primarily of timber and carbon sequestration, although they also aected biodiversity levels to a much
lower degree. These productivities were calculated according to equations and values from (Hgglund and Lundmark, 1987) and (Johansson et al.,
2013), using tree species, mean forest height and age, and total wood volume data from SLU Forest Map (SLU, 2010) for the year 2010 at 2020 m
resolution for Sweden. Productivity values were calculated at a 1 km2 resolution by averaging values from 400 m2 cells. Baseline grass productivities
were obtained by extracting the data for the year 2010 from the LPJ-GUESS grass NPP projections from the EC-EARTH model, RCP 4.5 (see section
B.7). For their use in CRAFTY-Sweden, baseline productivities were normalised for each forest type and grassland to values [0,1]. Table B.2 presents
nal mean, maximum and minimum baseline productivity levels for all forest types and grassland, and for the four larger administrative divisions
used by the Swedish Forest Agency.
Given the relative importance of transport infrastructure and urban areas in land-use activities, we generated an infrastructure indicator that
indicated proximity to transport networks and central markets (generally urban areas). This indicator is particularly relevant to the provision of
timber and recreation. Timber needs to be transported to areas where it can be processed and sold. Recreation in Sweden is most demanded close to
urban areas, as that is where the majority of the Swedish population lives, what makes peri-urban forests (currently approximately 15% of forest
cover) very important for recreation (Olsson, 2014). To calculate proximity to transport networks, we used data on the locations throughout Sweden

Table B.2
Mean, maximum and minimum (normalised) baseline productivity levels for all forest types and grassland, and for the four larger Swedish administrative divisions.

Pine Spruce Boreal Br. Nemoral Br. Grassland

Upper Norrland Max 0.57 0.59 0.47 0.15 0.52

Mean 0.21 0.16 0.07 0.06 0.24
Min 0.03 0.02 < 0.01 0.00 0.00

Lower Norrland Max 0.73 0.76 0.66 0.36 0.67

Mean 0.30 0.29 0.18 0.09 0.46
Min 0.02 0.02 0.12 0.03 0.00

Svealand Max 0.75 0.91 1.00 0.39 0.83

Mean 0.42 0.51 0.31 0.12 0.67
Min 0.02 0.03 < 0.01 0.03 0.01

Gtaland Max 1.00 1.00 0.85 1.00 1.00

Mean 0.48 0.65 0.35 0.26 0.78
Min 0.09 0.09 0.02 < 0.01 0.51

199
V. Blanco et al. Ecosystem Services 23 (2017) 174208

of roads (UNECE 2015a), railways (UNECE 2015b), and waterways (EEA, 2015), and calculated the Euclidian distance to each of them from every
pixel in Sweden at a 1 km2 resolution. We then normalised the distance values for the three resulting layers so that the largest distances would result
in the lowest proximity index values, using the equation:
PI = 1D / MaxD
where PI is the proximity index, D is the distance to the nearest road, railway or waterway, and MaxD is the maximum distance to these calculated
for Sweden. Assuming equal importance of the three modes of transport, we then summed the values for each pixel and divided the result by three to
obtain an overall index for proximity to transport networks.
To calculate the proximity to central markets we used data on accessibility (i.e. travel times) to the nearest town with a population greater than
50,000 in the year 2000 using land (road/o road) or water (navigable river, lake and ocean) based travel (Nelson, 2008). We normalised these data
using the above equation, where D was the distance to the nearest town. Subsequently, normalised values of proximity to central markets and
proximity to transport networks (assuming equal importance of both) were added and divided by two to achieve the nal transport infrastructure
index.

B.3 Baseline land-use and land owner distribution

Land-use map
We used CORINE land-use data for 2006 at a resolution of 100 m2 to determine the location of the more distinct land-use types (e.g. agriculture,
forest, urban), and SLU Forest Map data (SLU, 2010) (which included percentage forest cover, percentage of pine, spruce, boreal broadleaf, and
nemoral broadleaf forest volume out of total forest volume, and mean forest age) for 2010 at a resolution of 1 km2 to determine the type of forest
corresponding to the cells identied as forest. Since forest cover in Sweden did not change signicantly between 2006 and 2010 (Swedish
Environmental Protection Agency, 2015), we used CORINE data to identify forest cover locations, and determined the type of forest within each
forest pixel using the SLU Forest Map data.
To do this, we rst reclassied the land-use categories in the CORINE dataset to: articial surfaces, agriculture, forest, semi-natural areas, open
spaces, wetlands and water bodies. We then aggregated the 100 m2 cells into 1 km2 cells using a majority algorithm, which determines the new
value of the cell based on the most popular values within the lter window.
To establish forest type locations, we rst multiplied percentage forest cover values with species fractions, to obtain the proportion of each forest
type in each cell. We then determined forest types based on these proportions. The following rules were dened to determine the forest type to
which a pixel would belong (forest type names refer to the proportion of forest types in a cell):

a. Non-forest: Pine + Spruce + Boreal Broadleaf + Nemoral Broadleaf < 50%

b. Pine: Pine 60%
c. Spruce: Spruce 60%
d. Boreal Broadleaf: Boreal Broadleaf 60%
e. Nemoral Broadleaf: Nemoral Broadleaf 60%, or Nemoral Broadleaf > all others
f. Pine-Spruce: Pine 6040% and Spruce 6040%, or Pine and Spruce > Boreal Broadleaf and Nemoral Broadleaf
g. Pine-Boreal Broadleaf (managed by passive owners): Pine 6040% and Boreal Broadleaf 6030%, or Pine and (Boreal Broadleaf 30%) >
Spruce and Nemoral Broadleaf
h. Pine-Boreal Broadleaf (managed by multi-objective owners): Pine 6040% and Boreal Broadleaf 3020%, or Pine and (Boreal Broadleaf < 30%)
> Spruce and Nemoral Broadleaf
i. Spruce-Boreal Broadleaf (managed by passive owners): Spruce 6040% and Boreal Broadleaf 6030%, or Spruce and (Boreal Broadleaf 30%)
> Pine and Nemoral Broadleaf
j. Spruce-Boreal Broadleaf (managed by multi-objective owners): Spruce 6040% and Boreal Broadleaf 3020%, or Spruce and (Boreal Broadleaf
< 30%) > Pine and Nemoral Broadleaf

The proportions for forest types f-j were established based on the proportions planted by forest owner types in ProdMod (see section B.4) based
on owner type preferences.
Protected areas were then superimposed on the resulting forest map, to transform cells falling within a protected area to protected land. Non-
productive areas, are also protected and cannot be managed for production (Swedish Forest Agency, 2014). These forests are on the least productive
forest land. We determined the location of non-productive forests by:

1. Assigning to each cell covered by forest (from the forest type dataset), excluding protected areas, the productivity value corresponding to the
forest type with the highest productivity for that cell. The type with the highest productivity was chosen to ensure that non-productive forest areas
was selected by considering the potential of every site to grow the most suitable forest type.
2. The Swedish Forest Agency (SFA) Statistics provide the proportion of forest area in each county that is productive and non-productive, so we
selected for each county the corresponding proportion of forest cells with the lowest productivity values and labeled them as non-productive.

The resulting forest cover mapped (derived from SLU Forest Map data) covered 71.9% of Sweden, while forest cover from the Corine data was
63%. Therefore, as SLU forest statistics (SLU, 2015) report Swedish forest cover to be 69%, we used the forest cover layer to identify forested cells,
and superimposed these onto CORINE data. Cells for which the CORINE forest cells did not overlap with our forest cells were converted into
unmanaged land. Fig. B.1 shows the nal land-use map.
Forest age was attributed according to the mean forest age value in the 1 km2 pixel as identied in the SLU Forest Map. While this dataset allows
to map forest age throughout the country, it showed to overestimate the area covered by middle aged forest, planted during 19301970, and
underestimate that of old forest and very young forest (see Fig. B.2), when compared to age frequency distribution in the National Forest Inventory
(Swedish Forest Agency, 2015). This is due to both the raw SLU Forest Map data and the method used to spatially aggregate the raw data to 1 km2
cells. Such a cell can represent approximately 10100 stands in southern Sweden and about 550 stands in northern Sweden, with potentially

200
V. Blanco et al. Ecosystem Services 23 (2017) 174208

Fig. B.1. Land use map generated for Sweden for the year 2010.

Fig. B.2. Area of productive forest land by planting periods in Sweden, from a) spatially aggregated SLU Forest Map data (used in this study) and b) National Forest Inventory data.

dierent stand ages. This means that the cells represent an average age and the information of distribution of ages between stands in the cells is lost.

Agent locations
Since non-productive forest cannot be managed for production, it was combined into one class with protected areas. Agricultural land and
(some) semi-natural vegetation were assigned to farmer agents according to the land-use intensity within those areas. We identied ve intensity
classes according to the Institute of Environmental Studies (2015): extensive arable, moderately intensive arable, very intensive arable, extensive
grassland, and intensive grassland. Agricultural land was distributed among the ve categories, while extensive grassland was located on semi-

201
V. Blanco et al. Ecosystem Services 23 (2017) 174208

natural vegetation (i.e. pasture) where they intersected. Extensive arable land was assigned to non-commercial cereal agents, while moderately and
very intensive arable was allocated to commercial cereal agents. This resulted in 4.9% of arable land being allocated to extensive cereal agents, which
closely resembles gures reported by the Swedish Board of Agriculture (2009) of 5.0% of the land dedicated to cereal production being used for
organic production. Extensive and intensive grassland were allocated to non-commercial and commercial livestock agents respectively.
The remaining semi-natural vegetation, unmanaged land, and wetlands were left unmanaged. Protected areas, non-productive forests, open
spaces with little or no vegetation, articial surfaces, and water bodies were not assigned to any agent, as these land covers are not expected to
undergo any important changes at the national scale.
Forest types in productive forest land were allocated to the dierent owner types for which production functions had been generated. This was
done using two dierent datasets with information about types of forest owners present within the dierent Swedish counties. These were: a) the
area of productive forest land by county and ownership classes for 2010, obtained from SFA Statistics, which classied ownership into state, state
owned companies, other public owners, private sector companies, individual owners, and other private owners; and b) the proportion of owners in
each county belonging to each cluster obtained from the survey data cluster analysis. Owners who had participated in the survey were considered to
correspond to individual owners in the SFA data.
As all owners surveyed owned productive forest, their replies, and consequently the cluster proles, were considered appropriate to determine
the proportion of productive forest land owned by each cluster. Since we were able to relate clusters to owner types identied in the theoretical
typology, we could use cluster proportions per county to determine the proportions of owner types within non-industrial private forest owners.
The state, state owned companies, and other public owners, because they promote sustainable forestry that balances its economic,
environmental and social aspects (Blanco et al. in preparation, b), were considered to be multi-objective owners. Private sector companies were
identied as productionists. As we found no information about other private owners that could help to decide on one or more owner types to dene
this group, we divided the proportion of land owned by them equally among the ve possible owner types.
Hence, we could calculate the proportion of owners within each SFA owner class per county and subsequently calculate the proportion of owner
types in each county. In the case of SFA individual owners, we calculated their distribution among the dierent owner types by multiplying the
proportion of this owner class in each county by the proportions of each owner type derived from the cluster analysis per county.
We then counted the number of cells of productive forest and of each forest type per county, and calculated the number of productive forest cells
per county corresponding to each owner type according to their proportions. Next, owner types were assigned to forest types. As pine and spruce
could only be assigned to productionists, pine-boreal broadleaf ( < 30%) and spruce-boreal broadleaf ( < 30%) could only be assigned to multi-
objective owners, and pine-boreal broadleaf (30%) and spruce-boreal broadleaf (30%) could only be assigned to passive owners, the number of
cells of these forest types were allocated to the corresponding owner types. The remaining number of agents of each owner type were allocated by
assigning probabilities of each owner type of managing each forest type in each county. These probabilities were a function of the likelihood of an
agent type occurring in a county (i.e. number of agent types per county) and the likelihood of them managing a certain type of forest. The latter was
dependent on the number of agent types that a forest type could be assigned to. For instance, if nemoral broadleaf could only be managed by
recreationalists, conservationists and passive owners, each agent type would have a 33.33% chance of being assigned to that type of forest.

B.4 Land owner service production

B.4.1 Timber
For timber production, ost , being the optimal production that an agent type is able to produce with optimal capital levels, corresponds to the
production (timber volume) for a given year within the growth period of the forest type (i.e. age-dependent production). This value is calculated by
CRAFTY using functions of maximum yearly standing volume that were generated after post-processing ProdMods model output of maximum
timber growth. Prodmod is an empirical stand growth and yield model (Eko, 1985). ProdMod simulations were run for all other owner types at a
latitude of 57, altitude of 100 m, with an understory of herbs and grasses, an initial basal area of 0.5 m2/ha, and breast-height age (i.e. elapsed time
since tree height exceeded breast height) of 1 year. Site index (i.e. dominant stand height at 100 years of forest age) values for pine and spruce
forests used in ProdMod simulations come from Swedish Forest Agency statistical data on Productive forest land area by county/region and site
index for pine forest and spruce forest for the period 20082012. Site index values for boreal and nemoral broadleaf forests were obtained by
applying a function (from Hgglund and Lundmark (1987) Handledning i bonitering med Skogshgsskolans boniteringssystem Del 2 diagram och
tabeller. Skogsstyrelsen, Jnkping, 70 pages) to go from site index for spruce to site index for beech. The site index for beech was assumed to be
valid for both nemoral and boreal broadleaf forests.
To generate maximum timber production equations we 1) plotted the timber volume data against the corresponding stand age, and 2) adjusted

Fig. B.3. Example of curves of timber growth through time for a productionist managing pine and spruce (dotted lines), generated from adjusting trendlines to ProdMod timber volume
data (triangles and squares). On years when thinning takes place two volume values can be seen (before and after thinning). R2 values show the goodness of t of the curves to the data.

202
V. Blanco et al. Ecosystem Services 23 (2017) 174208

Fig. B.4. Equation used to estimate coarse woody debris accumulation in forests at a certain forest age.

trend lines in MS Excel to the generated curve. Among the possible trend lines we chose the one that tted the curve best (i.e. with the highest R2).
Possible trend line functions to t the data were exponential, linear, logarithmic, polynomial, and power. Polynomial equations gave the best t for
all timber growth datasets (Fig. B.3). To achieve the best possible t, degree two polynomial equations with R2 < 0.98 were substituted for degree
three polynomial equations if the latter had a better t to the data. A similar procedure was used to adjust best tting functions to above-ground
carbon growth data.
Capitals used in the production of timber were forest (type) productivities and transportation infrastructure. We attributed infrastructure
sensitivity values 0.30.5 according to the socio-demographic attributes (i.e. property size) of the owner types, and to their baseline distribution
throughout Sweden, so that if an owner concentrated for instance in the south, where highest infrastructure levels occur, they was assumed to be
more sensitive to infrastructure.

B.4.2 Carbon sequestration

For sequestered carbon, we calculated maximum productions functions of above ground carbon (excluding the stump) to be used in the Cobb
Douglas equation following a similar procedure as that for timber production. This could be done for all forest types except nemoral broad-leaved
forests using above ground biomass (excluding the stump) results from ProdMod simulations, and multiplying them by the carbon fraction of
aboveground forest biomass given by the IPCC (2006), which is dierent for conifers (0.51) and broad-leaved forests (0.48).
Because ProdMod does not generate biomass data for beech and oak, we could not use the simulation results to inform above ground biomass of
nemoral broad-leaved forests. Instead, to calculate this biomass, we used ProdMods ve-yearly volume per hectare results (which refer to the entire
stem cone above the bark), and added them to estimations of branch volume (down to a diameter of 5 cm) per hectare for both beech and oak, which
we calculated using the following equations (Hagberg and Matrn 1975):
Beech branch volume (m3) = (0.02080*D 2*H 0.24212*D*H 0.0003486*D 2*H2 )*0.001*S
Oak branch volume (m3) = (0.02813*D 2*H 0.3178*D*H 0.0006658*D 2*H2 )*0.001*S

where D is the diameter over the bark at breast height (cm), H is the height from the ground (m), and S is the number of stems per hectare.
We then multiplied the resulting above ground volumes of beech and oak by their corresponding wood densities, 580 and 720 kg dry matter/m3
fresh volume respectively (Cienciala et al., 2005; IPCC, 2006; Sdra, 2013) to obtain above ground biomass. Sdra (2013) suggests that oak density
may range in Sweden from between 690 and 760 kg/m3 (at 15% moisture content). As oak density decreases with increasing growth rate, assuming
a linear relationship, we chose a medium density value, given that we are simulating the fastest possible growth (i.e. maximum potential production
per unit of time) for nemoral broadleaf forests managed by low intensity forest owner types (i.e. recreationalist, conservationist, and passive)
(Blanco et al., 2015). Aggregating the biomass of both species results in nemoral broad-leaved forest biomass.
For all owner functional types managing more than one type of forest (e.g. productionist pine-spruce), the carbon content values of both forest
types were aggregated for each time step to obtain the total maximum carbon content that could be generated by the agent type.
Since carbon is incorporated by a tree as it grows in volume, and we conrmed that timber volume production and above ground carbon
production are strongly linearly correlated (tested with ProdMod simulation results for dierent agent types) (i.e. they grow at the same rate with
time), we could assume that, just like timber volume production, above ground carbon production could be underpinned by forest type dependent
productivities. Hence, we could use forest land productivity with maximum carbon production functions in the Cobb Douglas function.
Additionally, carbon production was recorded as negative when the forest was felled to reect carbon removal.

B.4.3 Biodiversity

The calculation of optimal forest biodiversity production considered forest age (Duncker et al., 2012b; Koskela et al., 2007; Marchetti, 2004),
tree diversity (Gamfeldt et al., 2013; Marchetti, 2004) and management practices undertaken by each owner type (e.g. woody debris removal),
which have an inuence on biodiversity (Blanco et al., 2015; Duncker et al., 2012a, 2012b). We chose these three forest attributes as indicators of
biodiversity because of the availability of baseline data and the possibility of updating the two capitals during the model simulations.
In a nationwide study in Sweden, Gamfeldt et al. (2013) reported understory plant species richness to be 31% greater in forests with ve rather
than one tree species. Hence, we estimated the increase in forest biodiversity (measured on a scale of 01) with each additional tree species to be
(31/4) 7.75%. Five classes resulted from separating forest types according to the number of tree species, with forests with ve or more tree species
being assigned a value of 1. One of the most important factors for sustaining a rich biodiversity in the boreal forest landscape is the presence of
deadwood (e.g. Berg et al., 1994; Jonsell et al., 1998; Siitonen, 2001). Hence, we used dead woody debris (i.e. deadwood with a minimum diameter
10 cm) as a proxy for biodiversity. We parameterised the eect of age on biodiversity by estimating dead woody debris volume as a logistic function
of age using the model of coarse woody debris accumulation from Sturtevant et al. (1997), while we normalised volume between 0 and the

203
V. Blanco et al. Ecosystem Services 23 (2017) 174208

Fig. B.5. Equation to estimate recreational value of a forest at a certain forest age.

maximum volume estimated at the age where the curve saturates (Fig. B.4).
Dierent owner types may have dierent impacts on biodiversity depending on their management practices (e.g. woody debris removal)
(Duncker et al., 2012b). The eect of management was considered for each owner type as a multiplying factor, and was parameterised using the
ndings of Duncker et al. (2012a, 2012b) as guidance.
The biodiversity production function looked like:
B = ME *NVt *TD
where, ME represents the eect of management, NVt is normalised coarse woody debris volume as a function of time, and TD is the eect of tree
diversity on understory plant species richness.
Finally, as biodiversity, and specically coarse woody debris, can be expected to be aected by forest productivity (Sturtevant et al., 1997), we
considered forest productivity as a capital aecting this service. Additionally, we attributed a conservative (i.e. low) capital weight of 0.06 to forest
productivity. For agents managing more than one forest type (e.g. pine-boreal broadleaf), the productivity of the most abundant forest type under its
management was chosen.

B.4.4 Recreation

Recreational value in Scandinavia is largely determined by the phase of development of a forest, but also by forest management practices, and,
only to a much smaller degree, by the type of tree species present (i.e. conifer vs broadleaf, and monoculture vs mixed) (Edwards et al., 2012).
Additionally, travel time and distance are well known to determine the accessibility of recreational areas. Recreation in forests was therefore
assumed to be a function of forest age, tree species type, accessibility (i.e. infrastructure capital), and management practices. To parameterise
production weights we used the results from the study by Edwards et al. (2012) to create an equation to calculate the maximum possible production
of an agent type with a forest of a certain age:
R = A*S*M
where A is the score assigned to a forest given its age, as estimated through the function shown in Fig. B.5, which was calculated from recreational
value scores assigned to phases of forest development (i.e. establishment (05 years), young (515 years), medium (1550 years) and adult (50+
years)). For modelling purposes, the result of this function is set to 0 for values where x=0. S and M are the scores given its tree species combination
(i.e. conifer, broad-leafed or mixed) or management strategy respectively (the highest score for each forest attribute being 1). Table B.3 shows S and
M scores.

Table B.3
Scores assigned to forest owner types given their species combination (S) and management strategy (M) to calculate their provision of
recreation.

Forest Agent Type S M

Productionist Pine 0.95 0.84

Productionist Spruce 0.95 0.84
Productionist Pine-Spruce 0.95 0.84
Productionist Boreal Br. 0.98 0.84
Multi-objective Pine-Spruce 0.95 0.96
Multi-objective Pine-Boreal Br. 1.00 0.96
Multi-objective Spruce-Boreal Br. 1.00 0.96
Multi-objective Boreal Br. 0.98 0.96
Recreationalist Pine-Spruce 0.95 1.00
Recreationalist Boreal Br. 0.98 1.00
Recreationalist Nemoral Br. 0.98 1.00
Conservationist Boreal Br. 0.98 0.94
Conservationist Nemoral Br. 0.98 0.94
Passive Pine-Boreal Br. 1.00 0.90
Passive Spruce-Boreal Br. 1.00 0.90
Passive Boreal Br. 1.00 0.90
Passive Nemoral Br. 0.98 0.90

204
V. Blanco et al. Ecosystem Services 23 (2017) 174208

As accessibility is primary in enabling recreation, assigned infrastructure capital sensitivities were between 0.6 and 1. Productionists and passive
owners were given levels of 1, under the assumption that because recreation is not among their objectives their forests will have little or no
recreational value if they are not easily accessible. In contrast, because recreation is recreationalists main objective, they were assumed to still be
able to provide this service even if they are further away, and were therefore assigned levels of 0.6. Multi-objective owners and conservationists were
assigned 0.8 for having recreation as a secondary objective.

B.5 Forest felling

The forest can be felled within a dened age distribution for each manager type. In Sweden, the time of felling is regulated by law for pine and
spruce. Minimum age at clear felling depends on the site index with some small dierences between counties. The site index is a measure of site
suitability to grow one or other tree type, so we can establish the minimum age of felling in relation to the forest productivity of the cell using linear
interpolation of rotation length between the highest and lowest productivities (Lagergren et al., 2012). Highest productivity values can correspond
to lowest minimum felling age, while lowest productivity values correspond to highest minimum felling age.
In Sweden, the minimum age for nal felling in stands where at least half the volume is made up of pine, at the most fertile sites, is 65 years, and
at sites with poorer soils it is 100 years. The minimum age for spruce at most fertile sites is 45 years, and at least fertile sites it is 90 years (Kunskap
and Direkt, 2015). Such regulation does not exist in Sweden for other tree species though. For beech, recommended rotation time is 100120 years,
and 120180 years for oak (Lf et al., 2009). As these two species make-up the nemoral broadleaf forest type, we dened the recommended rotation
times for the type to be the mean of the maximum and minimum ages for the period for each species (i.e. 110150 years). For birch the
recommended rotation time is 4060 years (Rytter et al., 2008). Having the highest and lowest minimum nal felling ages for pine and spruce and
recommended felling ages for broadleaf species, we can plot them against the maximum and minimum productivity capital values (i.e. 01) present
at baseline and subsequently interpolate a linear trend line between the two points. The resulting equations for pine and spruce dene the minimum
age of felling for each cell, beyond which an agent with an interest in timber production needs to go before felling its forest, while for broadleaf
species the equation establishes an approximate age at which the forest should be felled. For recreationalists, conservationists and passive owners
managing broadleaf forests, as they are not primarily interested in timber production, we dened felling ages beyond the recommended rotation
period.
Agent types with more than one tree species (e.g. productionist pine-spruce) present two dierent minimum clear felling ages, even though a cell
in CRAFTY-Sweden is entirely cleared at the time of nal felling. To have only one felling age for each cell, we chose the minimum felling age
function corresponding to the forest type that would generate the most timber volume according to ProdMod results, while ensuring that this
remained consistent with owner type objectives.

6.1. B.6 Climate change impact on productivities

We rst calculated yearly timber volume growth (m3 ha-1 yr-1) using LPJ-Guess projections for each forest type for each GCM-RCM ensemble
and each RCP. Each data value is given for a 50 km2 cell, and is an average for all forest patches with all ages represented within the cell. This cell
resolution is sucient because we are interested in the changes in its values associated to climate change, which is a large scale phenomenon that
would not cause signicant dierences at smaller scales (e.g. 1 km2). To calculate yearly volume growth we applied the equation:
G = Vt Vt-1+ H 10/(0.40.5).
where V is standing timber volume (m3 over bark ha-1, m3 over bark being the volume of the entire stem cone including bark) and H is the
harvested carbon (kg C m-2). The 10 results from the unit conversion 10,000 m2/ha 1 tonne/1000 kg, 0.4 is wood density (tonne dry wood/m3
wood), and 0.5 is the carbon fraction of wood biomass (tonne C/tonne dry wood). This calculation of volume growth assumes that the wood density
of all species is the same, which is not true. Therefore, to be able to compare the dierent forest types, volume growth was adjusted by a factor of 4/5
for boreal broadleaf and 2/3 for nemoral broadleaf.
We then calculated a linear equation for each cell for the dierent forest types with yearly production values, and also for grass using grass net
primary productivity (NPP), which provides the yearly increase in the capital attributable to climate change. We did this by creating an equation of
change in timber production and grass NPP with time for a past period (19512000) and another one for the future (20102100). Yearly increase
values for each cell where then downscaled to 1 km2 by dividing the 50 km2 cells, while maintaining the same values. Because the grid of values did
not cover the whole of Sweden due to the original coarse data resolution, we assigned to locations with no growth/ NPP values the values of the
nearest neighbours.
We normalised the baseline grassland productivity values between 0 and 1, where 0 is the minimum possible productivity in Sweden and 1 is the
maximum productivity. Since the values of forest productivities were not available for the whole of Sweden, we assigned values to locations where
they were missing by attributing to them the mean value of the ten nearest neighbours. Because nemoral broadleaf baseline productivity was only
available for the southernmost part of the country, we attributed values to locations missing productivities by assuming decreasing south-north
(because nemoral forests are adapted to nemoral climate, which occurs in the south of Sweden) and east-west (because mountains cover part of the
west of Sweden) gradients, while also attempting to assign new values in the south that are somewhat close to the existing ones. We also normalised
the growth and NPP rates of change by the maximum baseline productivity value of the corresponding forest type and of grass. Finally, each years
forest productivity in a cell was calculated using the equation:
Pt = Pt0 + (Y *R )
where Pt0 is the normalised baseline productivity, Y is the number of the years into the simulation (e.g. if 2012, Y=2), and R is the normalised yearly
increase. This may give us in some cases values > 1, which would be reasonable because the capital is multiplied by the production weight in the
Cobb Douglas function. As the production weight reects the maximum production that an agent can achieve in the present (i.e. it is not climate
sensitive), in the future, in a cell with a high productivity value the impact of climate change on the productivity could allow it to go beyond its
maximum production capacity in 2010.

205
V. Blanco et al. Ecosystem Services 23 (2017) 174208

Table B.4
SSP scenario descriptions in relation to ecosystem service demands, importance to society of meeting service demand levels, and giving-up and giving-in thresholds (reflecting
willingness to sell land and tolerance to competition respectively) (from Blanco et al. in preparation, a).

Reference Demands for all services remain unchanged through time.

Higher importance of meeting demands for pine and spruce timber than for boreal and nemoral broadleaf timber. Medium importance for
all other services.
More prot-oriented owner types, being more sensitive to benet values, have higher giving-up thresholds and lower giving-in thresholds.
Farmland cannot displace forestry.

SSP1 Demands for timber and carbon sequestration grow until the end of the century. Demands for biodiversity and recreation are stable until
Sustainability 2050 and grow thereafter. Cereal and meat demands grow until the middle of the century and slowly decrease afterwards.
Farmland cannot displace forestry.
Compared to the Reference, higher importance for carbon sequestration, biodiversity and recreation, and lower for cereal and meat.
Similar giving-up and giving-in thresholds as under the Reference.

SSP3 Demands for timber, carbon sequestration, biodiversity and recreation decrease, and cereal and meat demands grow, until the end of the
Regional Rivalry century.
Compared to the Reference, lower importance of meeting demands for carbon sequestration, biodiversity and recreation, and higher for
cereal and meat.
Lower giving-up and higher giving-in thresholds than under the Reference
Farmland can displace forestry.

SSP4 Demands for timber and carbon sequestration grow (less than under SSP1-RCP4.5) until 2050, and decline thereafter. Demands for
Inequality biodiversity and recreation decrease throughout. Demands for cereal and meat grow until the middle of the century (less than under SSP1-
RCP4.5 for cereal, but equally for meat) and stabilise thereafter.
Compared to the Reference, lower importance of meeting demands for timber, carbon sequestration, biodiversity and recreation, and
higher for cereal.
Similar giving-up and giving-in thresholds as under SSP3-RCP4.5.
Farmland cannot displace forestry.

SSP5 Demands for timber, carbon sequestration, decline during the first half of the century and grow back to initial levels afterwards.
Fossil-fuelled development Biodiversity and recreational demands decrease over the century, but at a higher rate during the first half. Demand for cereal grows
substantially during the first half of the century and remains relatively stable thereafter. Meat demands more than doubles in the course of
the first sixty years and slowly declines thereafter.
Similar to SSP3-RCP4.5, but higher importance of meeting demands for meat.
Similar giving-up and giving-in thresholds as under the Reference.
Farmland can displace forestry.

B.7 Scenario analysis

SSP narratives
Table B.4 presents a description for each of the SSPs used in the SSP-RCP scenario simulations.

Demand projections
Baseline (i.e. for 2010) demands for timber, cereal and meat were assumed to be equal to observed production in 2010. Timber production for
the dierent forest types were calculated using data about annual gross fellings and industrial consumption of roundwood by species (Swedish
Forest Agency, 2015). Cereal and meat production were sourced directly from (FAO, 2015). As no reliable empirical data to establish baseline
demands were found for sequestered carbon, biodiversity and recreation, these were assumed to be equal to the modelled aggregate supply in 2010.
To estimate changes in demands through time for the dierent scenarios for timber, carbon, cereals and meat we used the IIASA SSP data
(IIASA, 2015). Projections of forest land cover until 2100, modelled under SSPs 1, 3 and 4 in combination with RCP 4.5, were used to calculate
decadal rates of change that were applied to the baseline demands for timber and carbon under the assumption that global forest cover area
positively correlates with timber and carbon demands. Because the IIASA projections were not calculated for RCP 8.5, we assumed forest cover to
follow the linear trend established between RCPs 4.5 and 6.0 for each year, which we extrapolated for RCP 8.5. The same process was used to
estimate changes in demands for cereal and meat, although we used IIASA modelled crop and livestock demands instead of forest cover to calculate
the rate of change in demands. Changes for biodiversity were estimated following the SSP storylines and with guidance from modelled global future
changes in species abundance from UNEP (2007). Rates of change in demands for recreation were assumed to be the same as those for biodiversity.

Additional simulation parameters

At each time step (i.e. year) 1% of available cells can be allocated to agents. Also, a number of potential new agents (equal to 0.6% of cells) search
cells each year, and each of these agents search 0.01% of available cells.

References 4, 550557. http://dx.doi.org/10.1038/NCLIMATE2250.

Alexandratos, N., Bruinsma, J., 2012. World Agriculture Towards 2030/2050: The 2012
Revision. FAO, Rome.
Andersson, M., Gong, P.C., 2010. Risk preferences, risk perceptions and timber harvest
decisions - An empirical study of nonindustrial private forest owners in northern
Sweden. For. Policy Econ. 12, 330339. http://dx.doi.org/10.1016/
j.forpol.2010.02.002.
Arneth, A., Brown, C., Rounsevell, M.D.A., 2014. Global models of human decision-
making for land-based mitigation and adaptation assessment nature. Clim. Change
Berg, A., Ehnstrom, B., Gustafsson, L., Hallingback, T., Jonsell, M., Weslien, J., 1994.
Threatened plant, animal and fungus species in Swedish forests - Distribution and
habitat associations. Conserv. Biol. 8, 718731. http://dx.doi.org/10.1046/j.1523-
1739.1994.08030718.x.
Blanco, V., Brown, C., Holzhauer, S., Vulturius, G., Rounsevell, M. (in preparation.a) The
importance of socio-ecological system dynamics in understanding adaptation to
global change in the forestry sector Submitted to J. Environ. Manag, 2016.
Blanco, V., Brown, C., Rounsevell, M.D.A., 2015. Characterising forest owners through
their objectives, attributes and management strategies. Eur. J. For. Res. 134,
10271041. http://dx.doi.org/10.1007/s10342-015-0907-x.

206
V. Blanco et al.
Blanco, V., Holzhauer, S., Brown, C., Rounsevell, M. (in preparation, b). A conceptual
model of environmental institutions and their adaptive actions in the Swedish
forestry sector Submitted to Environ. Sci. Policy, 2016.
Blennow, K., Persson, J., Tome, M., Hanewinkel, M., 2012. Climate change: believing
and seeing implies adapting. PLoS One 7. http://dx.doi.org/10.1371/
journal.pone.0050182.
Blennow, K., Persson, J., Wallin, A., Vareman, N., Persson, E., 2014. Understanding risk
in forest ecosystem services: implications for eective risk management,
communication and planning. Forestry 87, 219228.
Boone, R.B., Galvin, K.A., BurnSilver, S.B., Thornton, P.K., Ojima, D.S., Jawson, J.R.,
2011. Using coupled simulation models to link pastoral decision making and
ecosystem services. Ecol. Soc. 16.
Boumans, R., Costanza, R., Farley, J., Wilson, M., Portela, R., Rotmans, J., Villa, F., et al.,
2002. Modeling the dynamics of the integrated earth system and the value of global
ecosystem services using the GUMBO model. Ecol. Econ. 41, 529560. http://
dx.doi.org/10.1016/S0921-8009(02)00098-8.
Brown, C., Alexander, P., Holzhauer, S., Rounsevell, M., in press. Behavioural models of
climate change adaptation and mitigation in land-based sectors WIREs Climate
Change, 2016.
Brown, C., Holzhauer, S., Metzger, M.J., Paterson, J.S., Rounsevell, M., 2016. Land
managers' behaviours modulate pathways to visions of future land systems. Reg.
Environ. Change, 115. http://dx.doi.org/10.1007/s10113-016-0999-y.
Brown, C., Murray-Rust, D., van Vliet, J., Alam, S.J., Verburg, P.H., Rounsevell, M.D.,
2014. Experiments in globalisation, food security and land use decision making.
PLoS One 9. http://dx.doi.org/10.1371/journal.pone.0114213.
Buonocore, E., Franzese, P.P., Ulgiati, S., 2012. Assessing the environmental
performance and sustainability of bioenergy production in Sweden: a life cycle
assessment perspective. Energy 37, 6978. http://dx.doi.org/10.1016/
j.energy.2011.07.032.
Carter, T.R., Rounsevell, M.D.A., Fronzek, S., Harrison, P.A., Holman, I.P., Pirttioja, N.
K., 2015. Integrated assessment approach. EU FP7 IMPRESSIONS Project
Deliverable D3.1.
Cienciala, E., Cerny, M., Apltauer, J., Exnerova, Z., 2005. Biomass functions applicable to
European beech. J. For. Sci. 51, 147154.
Claesson, S., Duvemo, K., Lundstrm, A., Wikberg, P.E., 2015. Skogliga
konsekvensanalyser. Skogsstyrelsen, Jnkping.
Costanza, R., et al., 1997. The value of the world's ecosystem services and natural capital.
Nature 387, 253260. http://dx.doi.org/10.1038/387253a0.
Douglas, P., 1976. The Cobb-Douglas production function once again: its history, its
testing, and some new empirical values. J. Polit. Econ. 84, 903915. http://
dx.doi.org/10.1086/260489.
Duncker, P.S., Barreiro, S.M., Hengeveld, G.M., Lind, T., Mason, W.L., Ambrozy, S.,
Spiecker, H., 2012a. Classication of forest management approaches: a new
conceptual framework and its applicability to European forestry. Ecol. Soc. 17,
(doi:51.10.5751/es-05262-170451).
Duncker, P.S., et al., 2012b. How forest management aects ecosystem services,
including timber production and economic return: synergies and trade-Os. Ecol.
Soc. 17, (doi:50.10.5751/es-05066-170450).
Edwards, D.M., et al., 2012. Public preferences across Europe for dierent forest stand
types as sites for recreation. Ecol. Soc. 17, (doi:27.10.5751/es-04520-170127).
EEA, 2014. Raster data on land cover for the CLC2006 inventory - Version 17. http://
www.eea.europa.eu/data-and-maps/data/corine-land-cover-2006-raster-3#tab-gis-
data, (accessed 04.03.15).
EEA, 2015. Nationally designated areas (CDDA). http://www.eea.europa.eu/data-and-
maps/data/nationally-designated-areas-national-cdda-10#tab-european-data,
(accessed 04.03.15).
Eko, P.M., 1985. En produktionsmodell for skog i Sverige, baserad pa bestand fran
riksskogstaxeringens provytor [A growth simulator for Swedish forests, based on
data from the National Forest Survey]. Swedish University of Agricultural Sciences
Engstrom, K., Olin, S., Rounsevell, M., Brogaard, S., van Vuuren, D., Alexander, P.,
Murray-Rust, D., Arneth, A., 2016. Assessing uncertainties in global cropland futures
using a conditional probabilistic modelling framework. Earth Syst. Dynam. 7,
893915. http://dx.doi.org/10.5194/esd-7-893-2016.
FAO, 2015. FAO Statistics Division - Production. http://faostat3.fao.org/browse/Q/*/E.(
accessed 15.12.15)
Felton, A., et al., 2016. How climate change adaptation and mitigation strategies can
threaten or enhance the biodiversity of production forests: insights from Sweden.
Biol. Conserv. 194, 1120. http://dx.doi.org/10.1016/j.biocon.2015.11.030.
Fulginiti, L.E., Perrin, R.K., 1998. Agricultural productivity in developing countries. Agr.
Econ. 19, 4551. http://dx.doi.org/10.1016/S0169-5150(98)00045-0.
Gamfeldt, L., et al., 2013. Higher levels of multiple ecosystem services are found in
forests with more tree species. Nat. Commun. 4, (doi:1340.10.1038/ncomms2328).
De Groot, R., et al., 2010. Integrating the ecological and economic dimensions in
biodiversity and ecosystem service valuation. In: Kumar, P. (Ed.), The Economics of
Ecosystems and Biodiversity: Ecological and Economic Foundations.. Earthscan,
London, Washington, 940.
Hagberg, E., Matrn, B., 1975. Tabeller fr kubering av ek och bok. Skogshgskolan, Inst.
f. skoglig matematisk statistik, Rapporter och Uppsatser 14, 118 s., Stockholm
Hgglund, B., Lundmark, J.E., 1987. Handledning I Bonitering med Skogshgsskolans
boniteringssystem, Del 2 Diagram och tabeller. Skogsstyrelsen, Jnkping.
Hayha, T., Franzese, P.P., Paletto, A., Fath, B.D., 2015. Assessing, valuing, and mapping
ecosystem services in Alpine forests Ecosystem Services 14:12-23 http://dx.doi.org/
10.1016/j.ecoser.2015.03.001.
IIASA, 2015. SSP Database (Shared Socioeconomic Pathways) - Version 1.0. https://
tntcat.iiasa.ac.at/SspDb/dsd?Action=htmlpage & page=about (accessed 02.11.15).
Ingemarson, F., Lindhagen, A., Eriksson, L., 2006. A typology of small-scale private
207
Ecosystem Services 23 (2017) 174208
forest owners in Sweden. Scand. J. For. Res. 21, 249259. http://dx.doi.org/
10.1080/02827580600662256.
Institute of Environmental Studies, 2015. Agricultural land use intensity data. VU
University Amsterdam. http://www.falw.vu.nl/en/research/earth-sciences/earth-
and-climate-cluster/research/environmental-geography/projects/agricultural-land-
use-intensity-data.aspx (accessed 05.09.15).
IPCC, 2006. 2006 IPCC Guidelines for National Greenhouse Gas Inventories, In:
Eggleston, H.S., Buendia, L., Miwa, K., Ngara, T., Tanabe, K. (Eds.). IGES, Japan.
Jacob, D., et al., 2014. EURO-CORDEX: new high-resolution climate change projections
for European impact research. Reg. Environ. Change 14, 563578. http://
dx.doi.org/10.1007/s10113-013-0499-2.
Johansson, U., Ek, P.M., Elfving, B., Johansson, T., Nilsson, U., 2013. Fakta Skog - Nya
hjdutvecklingskurvor fr bonitering. Swedish University of Agricultural Sciences,
Linkping.
Jones, C., Giorgi, F., Asrar, G., 2011. CLIVAR Exchanges - The Coordinated Regional
Downscaling Experiment: CORDEX, An international downscaling link to CMIP5 vol
16.
Jonsell, M., Weslien, J., Ehnstrom, B., 1998. Substrate requirements of red-listed
saproxylic invertebrates in Sweden. Biodivers. Conserv. 7, 749764. http://
dx.doi.org/10.1023/a:1008888319031.
Jonsson, A.M., Lagergren, F., Smith, B., 2015. Forest management facing climate change
- an ecosystem model analysis of adaptation strategies. Mitig. Adapt. Strateg. Glob.
Change 20, 201220. http://dx.doi.org/10.1007/s11027-013-9487-6.
Karali, E., Brunner, B., Doherty, R., Hersperger, A.M., Rounsevell, M.D.A., 2013. The
eect of farmer attitudes and objectives on the heterogeneity of farm attributes and
management in switzerland. Hum. Ecol. 41, 915926. http://dx.doi.org/10.1007/
s10745-013-9612-x.
Kjaer, E.D., Lobo, A., Myking, T., 2014. The role of exotic tree species in Nordic forestry.
Scand. J. For. Res. 29, 323332. http://dx.doi.org/10.1080/
02827581.2014.926098.
Kok, K., Pedde, S., Jger, J., Harrison, P., 2015. European Shared Socioeconomic
Pathways vol 2015. EU FP7 IMPRESSIONS Report.
Koskela, E., Ollikainen, M., Pukkala, T., 2007. Biodiversity conservation in commercial
boreal forestry: the optimal rotation age and retention tree volume. For. Sci. 53,
443452.
Kunskap, Direkt, 2015. Minimilder. http://www.kunskapdirekt.se/sv/KunskapDirekt/
Avverka/Grunder/Skogsvardslagen/Minimialder/ (accessed 03.08.15).
Lagergren, F., Jonsson, A.M., Blennow, K., Smith, B., 2012. Implementing storm damage
in a dynamic vegetation model for regional applications in Sweden. Ecol. Model. 247,
7182. http://dx.doi.org/10.1016/j.ecolmodel.2012.08.011.
Lavorel, S., Daz, S., Cornelissen, J.H.C., Garnier, E., Harrison, S.P., McIntyre, S., Pausas,
J.G., et al., 2007. Terrestrial ecosystems in a changing world. In J. G. Canadell, D. E.
Pataki, & L. F. Pitelka (Eds.), (pp. 149-164). Berlin, Heidelberg: Springer Berlin
Heidelberg. http://dx.doi.org/10.1007/978-3-540-32730-1_13.
Lidskog, R., Sjdin, D., 2014. Why do forest owners fail to heed warnings? Conicting
risk evaluations made by the Swedish forest agency and forest owners. Scand. J. For.
Res. 29, 275282. http://dx.doi.org/10.1080/02827581.2014.910268.
Lf, M., Mller-Madsen, E., Rytter, L., 2009. Sktsel av dellvskog. Skogsstyrelsen.
Marchetti, M., 2004. Monitoring and Indicators of Forest Biodiversity in Europe - From
Ideas to Operationality. European Forest Institute, Saarijarvi, Finland.
Martin, W., Mitra, D., 2001. Productivity growth and convergence in agriculture versus
manufacturing. Econ. Dev. Cult. Change 49, 403422. http://dx.doi.org/10.1086/
452509.
Matthews, R.B., Gilbert, N.G., Roach, A., Polhill, J.G., Gotts, N.M., 2007. Agent-based
land-use models: a review of applications. Landsc. Ecol. 22, 14471459. http://
dx.doi.org/10.1007/s10980-007-9135-1.
MEA, 2005. Ecosystems and human well-being: Synthesis. Washington, DC
Metzger, M.J., Schroter, D., Leemans, R., Cramer, W., 2008. A spatially explicit and
quantitative vulnerability assessment of ecosystem service change in Europe. Reg.
Environ. Change 8, 91107. http://dx.doi.org/10.1007/s10113-008-0044-x.
Murray-Rust, D., Brown, C., van Vliet, J., Alam, S.J., Robinson, D.T., Verburg, P.H.,
Rounsevell, M., 2014. Combining agent functional types, capitals and services to
model land use dynamics. Environ. Model. Softw. 59, 187201. http://dx.doi.org/
10.1016/j.envsoft.2014.05.019.
Murray-Rust, D., Dendoncker, N., Dawson, T.P., Acosta-Michlik, L., Karali, E.,
Rounsevell, M., 2011. Conceptualising the analysis of socio-ecological systems
through ecosystem services and agent-based modelling. J. Land Use Sci. 6, 8399.
Nelson, A., 2008. Estimated travel time to the nearest city of 50,000 or more people in
year 2000. Global Environment Monitoring Unit - Joint Research Centre of the
European Commission, Ispra Italy http://bioval.jrc.ec.europa.eu/products/gam/
(accessed 07.06.15).
O'Neill, B.C., et al., 2014. A new scenario framework for climate change research: the
concept of shared socioeconomic pathways. Clim. Change 122, 387400. http://
dx.doi.org/10.1007/s10584-013-0905-2.
Olsson, O.,, 2014. Out of the Wild - Studies on the forest as a recreational resource for
urban residents. Ume University
R Development Core Team, 2008. R: A language and environment for statistical
computing. R Foundation for Statistical Computing. http://www.R-project.org
(accessed 02.11.15).
Rammer, W., Seidl, R., 2015. Coupling human and natural systems: simulating adaptive
management agents in dynamically changing forest landscapes. Glob. Environ.
Change-Hum. Policy Dimens. 35, 475485. http://dx.doi.org/10.1016/
j.gloenvcha.2015.10.003.
Roberge, J.M., et al., 2016. Socio-ecological implications of modifying rotation lengths in
forestry. Ambio 45, S109S123. http://dx.doi.org/10.1007/s13280-015-0747-4.
Robinson, D.T., Brown, D.G., Parker, D.C., Schreinemachers, P., Janssen, M.A., Huigen,
V. Blanco et al.
M., Wittmer, H., et al., 2007. Comparison of empirical methods for building agent-
based models in land use science. J. Land Use Sci. 2, 3155. http://dx.doi.org/
10.1080/17474230701201349.
Rounsevell, M.D.A., Robinson, D.T., Murray-Rust, D., 2012. From actors to agents in
socio-ecological systems models. Philos. Trans. R. Soc. B 367, 259269. http://
dx.doi.org/10.1098/rstb.2011.0187.
Rytter, L., Karlsson, A., Karlsson, M., Stener, L.-G., 2008. Sktsel av bjrk, al och asp.
Skogsstyrelsen.
Satake, A., Leslie, H.M., Iwasa, Y., Levin, S.A., 2007. Coupled ecological-social dynamics
in a forested landscape: spatial interactions and information ow. J. Theor. Biol.
246, 695707. http://dx.doi.org/10.1016/j.jtbi.2007.01.014.
Schroter, D., et al., 2005. Ecosystem service supply and vulnerability to global change in
Europe. Science 310, 13331337. http://dx.doi.org/10.1126/science.1115233.
Scoones, I., 2014. Sustainable Rural Livelihoods: A Framework for Analysis. IDS http://
mobile.opendocs.ids.ac.uk/opendocs/handle/123456789/3390#.VxpMnEZx1hA
(accessed 15.01.15).
Siitonen, J., 2001. Forest management, coarse woody debris and saproxylic organisms:
Fennoscandian boreal forests as an example. Ecol. Bull. 49, 1141.
SLU, 2010. SLU Forest Map. ftp://salix.slu.se/download/skogskarta.( accessed 03.03.
14)
SLU, 2015. Current data about the Swedish forests from the Swedish National Forest
Inventory. http://www.slu.se/en/webbtjanster-miljoanalys/forest-statistics/
skogsdata/ (accessed 08.04.16).
Smith, B., Prentice, I.C., Sykes, M.T., 2001. Representation of vegetation dynamics in the
modelling of terrestrial ecosystems: comparing two contrasting approaches within
European climate space. Glob. Ecol. Biogeogr. 10, 621637. http://dx.doi.org/
10.1046/j.1466-822X.2001.t01-1-00256.x.
Sdra, 2013. Timber and lumber products from Swedish oak. http://www.sodra.com/
en/SodraTimberWorld/Our-products/Timber-and-lumber-products-from-birch-
oak-and-alder/Timber-and-lumber-products-from-Swedish-oak/ (accessed 13.07.
15).
Sohel, M.S.I., Mukul, S.A., Burkhard, B., 2015. Landscape's capacities to supply
ecosystem services in Bangladesh: a mapping assessment for Lawachara National
Park. Ecosyst. Serv. 12, 128135. http://dx.doi.org/10.1016/j.ecoser.2014.11.015.
Soja, A.J., et al., 2007. Climate-induced boreal forest change: predictions versus current
observations. Glob. Planet Change 56, 274296. http://dx.doi.org/10.1016/
j.glopacha.2006.07.028.
Sturtevant, B.R., Bissonette, J.A., Long, J.N., Roberts, D.W., 1997. Coarse woody debris
as a function of age, stand structure, and disturbance in boreal Newfoundland. Ecol.
Appl. 7, 702712.
Swedish Board of Agriculture, 2009. Facts about Swedish Agriculture. Jnkping,
Sweden.
Swedish Environmental Protection Agency, 2016. Milestone targets. http://www.
swedishepa.se/Environmental-objectives-and-cooperation/Swedens-environmental-
objectives/Milestone-targets/ (accessed 04.07.16).
Swedish Environmental Protection Agency, 2015. National inventory report Sweden
2015: Greenhouse gas emission inventories 19902013. Naturvrdsverket,
Stockholm, Sweden
Swedish Forest Agency, 2014. The Swedish Forestry Act. http://www.skogsstyrelsen.se/
208
Ecosystem Services 23 (2017) 174208
en/forestry/The-Forestry-Act/The-Forestry-Act/ (accessed 15.09.14).
Swedish Forest Agency, 2015. Statistics - Subject Areas. http://www.skogsstyrelsen.se/
en/AUTHORITY/Statistics/Subject-Areas/ (accessed 04 03.15).
Thompson, J.R., Foster, D.R., Scheller, R., Kittredge, D., 2011. The inuence of land use
and climate change on forest biomass and composition in Massachusetts. USA Ecol.
Appl. 21, 24252444.
Tilman, D., et al., 2001. Forecasting agriculturally driven global environmental change.
Science 292, 281284. http://dx.doi.org/10.1126/science.1057544.
Tzilivakis, J., Warner, D.J., Green, A., Lewis, K.A., 2015. Adapting to climate change:
assessing the vulnerability of ecosystem services in Europe in the context of rural
development. Mitig. Adapt. Strateg. Glob. Change 20, 547572. http://dx.doi.org/
10.1007/s11027-013-9507-6.
UNECE, 2015a. E-Road Trac Census in Europe (2005) Available at http://www.unece.
org/trans/main/wp6/e-roads_census_2005.html (accessed 15.10.15)
UNECE, 2015b. E-Rail Trac Census in Europe (2005) Available at http://www.unece.
org/trans/main/wp6/e-rails_census_2005.html (accessed 15.10.15)
UNEP, 2007. Global Environment Outlook 4: Environment for Development. United
Nations Environment Programme, Valletta, Malta.
Van Asselen, S., Verburg, P.H., 2013a. Land cover change or land-use intensication:
simulating land system change with a global-scale land change model. Glob. Change
Biol. 19, 36483667. http://dx.doi.org/10.1111/gcb.12331.
Valbuena, D., Verburg, P.H., Bregt, A.K., Ligtenberg, A., 2010. An agent-based approach
to model land-use change at a regional scale. Landsc. Ecol. 25, 185199. http://
dx.doi.org/10.1007/s10980-009-9380-6.
Valbuena, D., Verburg, P.H., Bregt, A.K., 2008. A method to dene a typology for agent-
based analysis in regional land-use research. Agric. Ecosyst. Environ. 128, 2736.
http://dx.doi.org/10.1016/j.agee.2008.04.015.
Verkerk, P.J., Mavsar, R., Giergiczny, M., Lindner, M., Edwards, D., Schelhaas, M.J.,
2014. Assessing impacts of intensied biomass production and biodiversity
protection on ecosystem services provided by European forests. Ecosyst. Serv. 9,
155165. http://dx.doi.org/10.1016/j.ecoser.2014.06.004.
Vulturius, G., Andr, K., Gerger Swartling, ., Brown, C., Rounsevell, M., Jnsson, A.,
Blanco, V., in preparation. Explaining engagement with climate change adaptation
among forest owners in Sweden. Submitted to Reg Environ Change, 2016.
van Vuuren, D.P., et al., 2011. The representative concentration pathways: an overview.
Clim. Change 109, 531. http://dx.doi.org/10.1007/s10584-011-0148-z.
van Vuuren, D.P., et al., 2014. A new scenario framework for climate change research:
scenario matrix architecture. Clim. Change 122, 373386. http://dx.doi.org/
10.1007/s10584-013-0906-1.
Van Asselen, S., Verburg, P.H., 2013b. Land cover change or land-use intensication:
simulating land system change with a global-scale land change model. Glob. Change
Biol. 19, 36483667. http://dx.doi.org/10.1111/gcb.12331.
Ward, J.H., Jr., 1963. Hierarchical grouping to optimize an objective function. J. Am.
Stat. Assoc. 58, 236244.
Zanchi, G., Pena, N., Bird, N., 2012. Is woody bioenergy carbon neutral? A comparative
assessment of emissions from consumption of woody bioenergy and fossil fuel. Glob.
Change Biol. Bioenergy 4, 761772. http://dx.doi.org/10.1111/j.1757-
1707.2011.01149.x.